JP2014235654A - Risk evaluation device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To quantize the degree of risk of project failure in a project in progress and find a chronological change in it, thereby detecting a symptom of project failure early.SOLUTION: Encoded project-in-progress information is generated by substituting a division value corresponding to a division to which belongs the value of each influential factor in project-in-progress information on the basis of division information of influential factor information. A division information lift value which is information indicating the degree of influence of the influential factor against success or failure of the project is calculated for timing information on each influential factor and each division value on the basis of encoded past project information and project-in-progress information. A risk score per timing information value about each project is calculated, for the encoded past project information or project-in-progress information, by using the timing information on each influential factor and the division information lift value for each division information.

Description

  Embodiments described herein relate to a risk evaluation apparatus.

  It is important to use various development management data (progress, cost, quality information, etc.) in order to quickly detect signs of problems and prevent project failures in large-scale and complicated development projects. Become. In addition, with the widespread use of various development support tools such as configuration management tools and defect management tools, it becomes possible to collect and accumulate development management data related to projects without placing an extra burden on the person in charge of development. I came.

  The potential risks of the project are not visible, the response tends to be delayed, and proactive management ahead of the risks is necessary. In a situation where changes are intense and uncertainty is high, there is a demand to quantitatively evaluate risks related to project failures, such as delays in delivery, cost overruns, and quality shortages, and to continuously monitor changes in risk predictions. there were.

  The following methods have been proposed for risk prediction.

(1) Prediction Model by Multiple Regression Analysis Multiple regression analysis is one of the most commonly used methods for building a prediction model in software development management. This is a technique for predicting the value of the objective variable itself when the value of the explanatory variable is obtained by directly obtaining the relational expression between the explanatory variable (input) and the objective variable (output) by multiple regression analysis.

(2) Confusion prediction system based on questionnaire survey Whether the project will eventually fall into a confused state based on the results of the questionnaire survey for a new project by learning the results of a questionnaire survey conducted on a past project using a Bayes classifier Is a method for determining

JP 2011-076411 A JP 2010-108404 A JP 2007-265141 A JP 2009-187288 A JP 2011-141684 A JP 2008-204313 A

The conventional risk prediction method as described above has the following problems.
The risk prediction method based on multiple regression analysis has a problem that it is easily affected by outliers and abnormal values. Building a highly accurate prediction model requires a large amount of data with high accuracy, but if there are missing values, the number of samples decreases. In general, software development management data is less accurate than hardware manufacturing process measurement data, and it is often difficult to construct a prediction model by multiple regression analysis.

  In the project confusion prediction system based on the questionnaire survey, the prediction is performed only at one time point after the questionnaire is aggregated, the update of the risk prediction value is not considered, and only the success / failure of the project is determined. It also does not calculate failure probabilities and does not provide a quantitative assessment. Also, only discrete value variables are input, and there is no mention of continuous value variables and their threshold determination methods.

  One embodiment of the present invention aims to provide a technique for quantifying the risk of project failure in an ongoing project, obtaining a time series change thereof, and detecting an early sign of project failure.

  Another embodiment of the present invention aims to provide a technique for calculating the probability of project failure at any point in time for an ongoing project.

  According to yet another embodiment of the present invention, by selecting the most influential factors and setting the classification information, it is possible to drastically reduce the labor of the prediction model construction work and improve the prediction accuracy of the project failure. The purpose is to provide technology.

The first embodiment of the present invention is proposed as a risk evaluation apparatus. The risk evaluation apparatus includes a first storage unit, a first processing unit, a second storage unit, and a second processing unit.
The first storage means distinguishes the past project information, which is information describing the values of a plurality of influencing factors for projects executed in the past, and the projects executed in the past into successful projects and failed projects. Project failure definition information that is information describing the criteria for the above, division information that divides the possible area of the influence factor value into multiple divisions, and timing information that indicates when the value of the influence factor can be acquired And influencing factor information that is information including

  The first processing means, based on the division information of the project failure definition information and the influence factor information, by replacing with the division value corresponding to the division to which the value of each influence factor of the past project information belongs, encoded past project information Based on the encoded past project information, the category information lift value, which is the information indicating the degree of the influence of the influence factor on the success or failure of the project, is calculated for each of the influence value of each influence factor.

The second storage means stores in-progress project information, which is information describing a plurality of influencing factor values for the in-progress project.
The second processing means generates encoded in-progress project information by substituting the segment value corresponding to the category to which each influencing factor value of the ongoing project information belongs based on the category information of the influencing factor information. Based on the encoded past project information and ongoing project information, the division information lift value, which is the information indicating the degree of influence of the influencing factors on the success or failure of the project, is calculated for each timing information and the dividing value of each influencing factor. Using the timing information of each influencing factor and the category information lift value for each category information for the encoded past project information or the ongoing project information, the risk score for each timing information value is calculated for each project. .

  The second embodiment of the present invention is proposed as a risk evaluation apparatus. This risk evaluation device obtains a regression formula of the project failure probability based on the success / failure determination value that is a value indicating the success or failure of the project in the encoded past project information and the risk score of the past project for each value of the timing information. The apparatus further includes a third processing unit that calculates a project failure probability for the ongoing project using the project failure probability regression equation.

  The third embodiment of the present invention is proposed as a risk evaluation apparatus. This risk evaluation apparatus generates encoded past project information from past project information using third storage means for storing influential factor candidate information which is a set of influential factor and its classification information candidates, and influential factor candidate information. If the category information lift value or the category information lift value is 1 or less, the expected category information lift value is calculated by calculating the expected category information lift value that is the sum of the expected value and the inverse of the category information lift value. And a fourth processing means for selecting an influence factor to be included in the influence factor information and its classification information based on the value, and generating influence factor information including the selected influence factor and the classification information.

Functional block diagram showing a configuration example of the risk evaluation device according to the first embodiment Diagram showing data structure example of past project information A diagram showing an example of the data structure of influencing factor information The figure which shows the data structural example of the encoding past project information The figure which shows the relation of the element for calculating division information lift value The figure which shows each value of c1, c2, c3, c4 in encoding past project information Another diagram showing the values of c1, c2, c3, and c4 in the encoded past project information Diagram showing data example of ongoing project information A diagram showing an example of the data structure of encoding ongoing project information The figure which shows the line graph which visualizes the time series change of the risk score in each timing information (T = 1, 2, 3, 4) Diagram showing time-series changes in risk score of ongoing projects (New1-New5) Flow chart showing an example of risk score calculation processing Functional block diagram showing a configuration example of the risk evaluation apparatus according to the second embodiment Diagram showing original data for deriving a project failure probability regression formula by logistic regression analysis Flow chart showing an example of project failure probability calculation processing Functional block diagram showing a configuration example of a risk evaluation apparatus according to the third embodiment The flowchart which showed the example of the main operation | movement of the risk evaluation apparatus which concerns on 3rd Embodiment. Functional block diagram showing an example of the configuration of a risk evaluation apparatus that combines the first, second, and third embodiments

Hereinafter, a risk evaluation apparatus according to an embodiment of the present invention will be described with reference to the drawings.
[0. Definition of terms]
Definitions of terms used in this specification are described.
(1) Project information “Project information” refers to development management data collected and recorded during or after project execution. “Project information” is expressed as a set of variable name and value pairs.

(2) Project failure “Project failure” means that the result of the project, which is determined by a certain evaluation standard, is in an undesirable state. Typical examples of “project failure” are poor quality, cost overrun, delay in delivery, etc.

(3) Influencing factors “Influencing factors” are development management data and attribute information that affect project failure. As attribute information, there are classification information and timing information.

(4) Classification information “Classification information” is one piece of attribute information of an influence factor, and is defined as a value range of a variable corresponding to the influence factor divided into several areas.

(5) Timing information “Timing information” is one of the attribute information of the influencing factors, and indicates the time (process, phase, milestone, etc.) when the value of the variable corresponding to the influencing factors can be obtained. The timing information is labeled as T = 1, 2, ..., m in order from the earliest.

(6) Category information lift value “Category information lift value” means that the probability that an influence factor value belongs to a category when the project fails is divided by the probability that the influence factor value belongs to the category when there is no condition. Value. The “category information lift value” indicates how much the information obtained by the influence factor contributes to the prediction of the project failure. If the category information lift value is “1”, it means that the influential factor does not contribute to the prediction of project failure.

(7) Risk score “Risk score” refers to a value obtained by calculating the posterior probability of project failure based on the Bayesian probability and taking the logarithm of the failure. It is considered that the probability of project failure increases as the risk score increases.

(8) Expected category information lift value “Expected category information lift value” means the expected value for the category information lift value or the inverse of the category information lift value (* when the category information lift value is 1 or less). Is the total value. It is used to determine the influence factor category information.

(9) Prediction Model A “prediction model” refers to a mathematical model that predicts future results (for example, delivery date, cost, quality, etc.) using data obtained in advance or during the project as input. In this specification, it refers to a predictive model of project failure risk.

(10) Bayesian statistics "Bayesian statistics" is a probability / statistical system based on Bayes' theorem and incorporating subjective probabilities. Contrast with probabilistic and statistical systems (called frequencyism) that do not allow subjective probability, such as speculative statistics.

(11) Naive Bayes classifier The “Naive Bayes classifier” is one of the applications of Bayesian statistics. Probabilistic classification is performed assuming event independence in the calculation of posterior probabilities. It is applied to spam mail filtering technology.

(12) Logistic regression analysis "Logistic regression analysis" is a kind of statistical regression model. Linear regression analysis predicts quantitative variables, whereas logistic regression analysis depends on binary qualitative variables. The occurrence probability is predicted using an explanatory variable as a variable.

[1. First Embodiment]
A first embodiment of the present invention will be described. The first embodiment is proposed as a risk evaluation apparatus.

  The risk evaluation apparatus according to the first embodiment quantifies the risk of project failure in an ongoing project as a “risk score” based on past project information composed of various measurement data, and changes its time series. Is a device that provides The risk assessment apparatus according to the first embodiment applies Bayesian statistics to obtain influential factors of project failure and the degree of impact (category information lift value) from past project data, and newly obtains in the ongoing project. Based on the obtained data, a risk score indicating the risk of project failure of this ongoing project is calculated. The risk evaluation apparatus according to the first embodiment can update the risk score according to the progress of the ongoing project.

  The risk evaluation apparatus is an information processing apparatus such as a computer or a workstation, and the information processing apparatus includes an arithmetic processing unit (CPU), a main memory (RAM), a read-only memory (ROM), and an input / output device (I / I). O) and, if necessary, a device having an external storage device such as a hard disk device.

[1.1. Configuration example of risk evaluation device according to first embodiment]
FIG. 1 is a functional block diagram showing a configuration example of the risk evaluation apparatus according to the first embodiment. Note that the components shown in the functional block diagram are the functions of the risk assessment device that are grouped by function and considered as a block, and the risk assessment device is a board, device, circuit, component, etc. corresponding to each component. It does not mean that it must have physical components. “Connected” means that data, information, instructions, etc. can be sent, received, delivered, etc. It is not limited. The same applies to the description of other functional block diagrams in this specification.

  The risk evaluation apparatus 1 includes an input information storage unit 30, a prediction model construction unit 10 connected to the input information storage unit 30, a risk score calculation unit 20 connected to the prediction model construction unit 10, and the risk score And an ongoing project information storage unit 40 connected to the calculation unit 20. The input information storage unit 30 corresponds to a first storage unit, the prediction model construction unit 10 corresponds to a first processing unit, and the ongoing project information storage unit 40 corresponds to a second storage unit.

[1.1.1. Input information storage unit]
The input information storage unit 30 has a function of storing data necessary for constructing a prediction model. The input information storage unit 30 stores past project information 200, project failure definition information 300, and influence factor information 400.

[1.1.1.1. Past project information]
The past project information 200 is information including a plurality of pieces of individual project information implemented in the past. The individual project information is various development management data collected and recorded during or after the execution of the project, and is represented as a set of variable name / value pairs of variables. The variable may be a continuous variable or a discrete variable. Some of the variable values may contain missing values.

  FIG. 2 shows a data configuration example of the past project information 200. The past project information 200 has a plurality of records 201 corresponding to individual project information. In the example shown in FIG. 2, there are 10 records 201 corresponding to 10 projects from the past project PJ1 to the past project PJ10.

  Each record 201 has a field for storing a variable value. In the example shown in FIG. 2, “spec review review indication density”, “design review indication density”, “design review efficiency”, “unit test bug density”, “joint test bug density”, and “post-shipment” "Bug density" is adopted. Each record 201 includes a “specification review indication density” storage field 211, a “design review indication density” storage field 212, a “design review efficiency” storage field 213, a “unit test bug density” storage field 214, and an “integration test”. It has a “bug density” storage field 215 and a “bag density after shipment” storage field 216.

  Each storage field 211 to 216 stores the value of the corresponding variable in the corresponding project. When the value of the variable is not obtained, information indicating “missing value” is stored.

[1.1.1.2. Project failure definition information]
The project failure definition information 300 is information defining a standard for classifying (distinguishing) a set of projects executed in the past into a “successful” project group and a “failed” project group.

  Typically, the project failure definition information 300 includes quantitative criteria (such as a post-shipment bug rate, a cost prediction ratio, days of delay, etc.) related to quality indicators, excess costs, delay in delivery date, and the like. (Threshold value) is set.

  The project failure definition information 300 corresponding to the example of the past project information 200 shown in FIG. 2 is “failure” if the post-shipment bug density value is 4.5 or more, and other values are “success”. In addition, “failure” and “success” each have a category value to be described later, and are “1” and “0”, respectively.

[1.1.1.3. Influence factor information]
The influence factor information 400 includes “category information” obtained by dividing an area where the value of the influence factor can be divided into several areas with respect to a set of influence factors which are variables that are thought to affect the success or failure of the project, It is composed of “timing information” which is information indicating the time (process, phase, milestone, etc.) at which the value of the influence factor can be acquired. The timing information may be T = 1, 2,..., M in order from the earliest (duplicate values).

  FIG. 3 shows a data configuration example of the influence factor information 400. The influence factor information 400 has one record 401 for each influence factor. Each record 401 stores information (variable name, variable number, variable ID, etc.) for specifying the influential factor, and an influential factor specifying information storage field 411 for storing the influential factor classification information. And a timing information storage field 413 for storing timing information of the influencing factors.

  The “classification information” is information indicating a classification (value range) that can be divided into a plurality of values that can be taken by the influential factor corresponding to the record. For example, in the example shown in FIG. 3, as the category information of the influence factor “use review indication density”, the first category (range) is “[MIN, 2.5)”, and the second category (range) Is “[2.5, MAX]”, and the third segment is recorded as “missing value”. Note that “[MIN, 2.5)” means a range between the minimum value and less than 2.5, and “[2.5, MAX]” means a range between 2.5 and the maximum value. doing.

  Further, each of the above-mentioned categories has a “classification value”. For example, the category value of the first category = 0, the category value of the second category = 1, and the category of the third category = NA. It is. This division value is used for encoding past project information described later.

  In the example shown in FIG. 3, “timing information” is expressed as “T = 1 (required specification process)” for the sake of explanation, but only the value is actually stored in the timing information storage field 413. It's enough.

[1.1.2. Prediction model construction department]
The prediction model construction unit 10 encodes the past project information 200 from the information stored in the input information storage unit 30 to generate encoded past project information. Based on the encoded past project information, the classification information lift It has a function to calculate a value.

[1.1.2.1. Encoded past project information]
Generation of encoded past project information will be described. The prediction model construction unit 10 encodes the past project information 200 based on the project failure definition information 300 and the classification information of the influence factor information 400. “Encoding” refers to replacing a value stored in the storage field of each record with a segment value. In the present embodiment, the values from the storage field 211 to the storage field 216 are replaced with the segment values corresponding to the values. The values from the storage field 211 to the storage field 215 are replaced with the partition values based on the partition information of the influence factor information 400. The value in the storage field 216 is replaced with a segment value based on the project failure definition information 300.

  FIG. 4 shows a data configuration example of the encoded past project information obtained as a result of encoding the past project information 200 shown in FIG. Similar to the past project information 200, the encoded past project information 500 has ten records 501 corresponding to ten past projects PJ1 to PJ10. Each record 501 includes a “specification review indication density” storage field 511, a “design review indication density” storage field 512, a “design review efficiency” storage field 513, a “unit test bug density” storage field 514, and a “combination test”. It has a “bug density” storage field 515 and a “post-shipment bug density” storage field 516.

  The prediction model construction unit 10 stores the segment values corresponding to the values stored in the corresponding storage fields 211 to 215 of the past project information 200 in the storage fields 511 to 515, respectively. For example, in the record 201 corresponding to the past project PJ1 of the past project information 200, the value stored in the specification review indication density storage field 211 is “0.5”. This value “0.5” is included in the first category (value range) “[MIN, 2.5)” described in the category information 412 of the record 401 corresponding to the specification review indication density in the influence factor information 400. . Since the classification value of the first classification (value range) “[MIN, 2.5)” is “0”, the prediction model construction unit 10 includes the record 501 corresponding to the past project PJ1 of the encoded past project information 500. The segment value “0” is stored in the specification review indication density storage field 511. Similarly, the segment values corresponding to the other storage fields 512 to 515 are stored.

  The prediction model construction unit 10 stores the corresponding segment value in the post-shipment bug density storage field 516. For example, in the record 201 corresponding to the past project PJ1 of the past project information 200, the value stored in the post-shipment bug density storage field 216 is “1.2”. Since this value “1.2” does not correspond to the failure condition described in the project failure definition information 300 that the value of the bug density after shipment is 4.5 or more, it is determined as “successful” and succeeded. The corresponding segment value “0” is stored in the post-shipment bug density storage field 516 in the record 501 corresponding to the past project PJ1 of the encoded past project information 500.

  The prediction model construction unit 10 determines corresponding segment values for all the storage fields 511 to 516 of all records 501 and stores the segment values in the storage fields 511 to 516.

  This is the encoded past project information 500 in a state where the storage of the segment values shown in FIG. 4 is completed. In the example illustrated in FIG. 4, the labels of the storage fields 511 to 516 include the corresponding influence factor (variable) names “specification review indication density”, “design review indication density”, “design review efficiency”, “unit test bug”. X1, X2a, X2b, X3, X4, and Y were used instead of “density”, “joint test bug density”, and “post-shipment bug density”. The number on the right side of X is the timing information value of the influencing factor. In addition, since “design review indication density” and “design review efficiency” are both timing information T = 2, subscripts a and b are further added for distinction. “Post-shipment bug density” indicating the success / failure of the project is indicated as Y. These notations are for explaining a calculation process described later, and are not intended to limit the data structure of the encoded past project information 500. In the example shown in FIG. 4, “NA” is a segment value corresponding to a missing value.

[1.1.2.2. Category information lift value]
The predictive model construction unit 10 sets the category information lift value, which is information indicating the degree of influence of the influence factor, on the success / failure of the project for each of the division values of each influence factor based on the encoded past project information 200 described above. calculate.

  In order to calculate the segment information lift value, the prediction model construction unit 10 refers to the encoded past project information 500 for the segment value k (0 ≦ k ≦ n) of the influence factor Xi (i is the value of timing information). Then, the following c1, c2, c3, c4 are obtained.

  “C1” indicates that the value (partition value) stored in the corresponding storage field for the influence factor Xi is “k”, and the value of the storage field 516 (Y) indicating the success / failure of the project is 0. This is the number of records 501.

  “C2” is the value stored in the storage field other than “k” and other than “NA” for the influence factor Xi, and the value of the storage field 516 (Y) indicating the success / failure of the project is This is the number of records 501 having a value “0” indicating success.

  “C3” is a value “1” indicating that the value stored in the corresponding storage field is “k” and the value of the storage field 516 (Y) indicating the success / failure of the project is “1” indicating the failure. Is the number of records 501.

  “C4” is the value stored in the corresponding storage field other than “k” and other than “NA” for the influencing factor Xi, and the value of the storage field 516 (Y) indicating the success / failure of the project is This is the number of records 501 that are one.

  FIG. 5 shows a table showing the relationship between Xi, k, c1, c2, c3, and c4.

  In the example of the encoded past project information 500 shown in FIG. 4, the prediction model construction unit 10 c1 and c2 for each of the division values “0” and “1” of the influence factors X1, X2a, X2b, X3, and X4. , c3, c4.

Next, the prediction model construction unit 10 calculates the segment information lift value “L_Xi_k” defined below.
The calculation formula of the classification information lift value “L_Xi_k” is shown below.

  In the above equation 1, the numerator is the probability that the value taken by the influencing factor Xi belongs to the category value k when the project fails, the denominator is the probability that the value taken by the influencing factor Xi belongs to the category value k without any condition. is there.

  A and b in the above equation 1 are correction parameters for avoiding extreme values being output when the number of data is small. Generally, a = 1, b = 2 (Laplace correction) and the like are used. If the number of data is sufficient, a = b = 0 (no correction) may be used.

  Assuming the independence of each influencing factor, the posterior probability of project failure matches the product of the prior probability and the classification information lift value when a = b = 0 (no correction) according to Bayes' theorem.

  The calculation example of a division | segmentation information lift value is shown. The division information lift value L_X1_0 for the influence factor = X1 and the division value = 0 in the encoded past project information 500 shown in FIG. 4 is as follows.

  FIG. 6 shows values of c1, c2, c3, and c4 in the encoded past project information 500 when the influence factor = X1 and the segment value = 0. That is, c1 = 4, c2 = 2, c3 = 1, and c4 = 2. This is applied to Equation 1 above.

(However, with Laplace correction: a = 1, b = 2)
That is, the division information lift value L_X1_0 = 0.7333 for the influence factor = X1 and the division value = 0 is obtained.

  The other example of calculation of a division | segmentation information lift value is shown. The division information lift value L_X1_1 for the influence factor = X1 and the division value = 1 in the encoded past project information 500 shown in FIG. 4 is as follows.

  FIG. 7 shows values of c1, c2, c3, and c4 in the encoded past project information 500 when the influence factor = X1 and the segment value = 0. That is, c1 = 2, c2 = 4, c3 = 2, and c4 = 1. This is applied to Equation 1 above.

(However, with Laplace correction: a = 1, b = 2)
That is, the segment information lift value L_X1_1 = 1.32 for the influence factor = X1 and the segment value = 1 is obtained.

  The prediction model construction unit 10 similarly calculates the division information lift value for each of the division values “0” and “1” of the other influencing factors X2a, X2b, X3, and X4, and outputs each division information lift value. .

  Above, description of the prediction model construction part 10 is complete | finished.

[1.1.3. Ongoing project information storage unit]
The ongoing project information storage unit 40 has a function of storing the ongoing project information 800. The ongoing project information 800 is a collection of one or more “in progress” (success / failure unconfirmed) project information. The ongoing project information 800 has the same data structure as the past project information 200 described above.

  FIG. 8 shows an example of ongoing project information data. The ongoing project information 800 includes a plurality of records 801 corresponding to individual ongoing project information. In the example shown in FIG. 8, there are five records 801 corresponding to five ongoing projects from the ongoing project NEW1 to the ongoing project NEW5.

  Each record 801 has a field for storing a value of a variable, like the past project information 200. In the example shown in FIG. 8, “specification review indication density”, “design review indication density”, “design review efficiency”, “unit test bug density”, “joint test bug density”, and “post-shipment” "Bug density" is adopted. Each record 801 includes a “specification review indication density” storage field 811, a “design review indication density” storage field 812, a “design review efficiency” storage field 813, a “unit test bug density” storage field 814, and a “combination test”. It has a “bug density” storage field 815 and a “post-shipment bug density” storage field 816.

  Each storage field 811 to 816 stores the value of the corresponding variable in the corresponding project. When the value of the variable is not obtained, information indicating “missing value” is stored. In addition, when verification / inspection or the like for the corresponding variable is actually performed, information indicating “not performed” is stored.

  The project information “after project completion” (after success / failure is confirmed) of the ongoing project can be included in the past project information 200 and used for prediction of the future ongoing project. .

[1.1.4. Risk score calculator]
The risk score calculation unit 20 has a function of calculating risk scores for past and ongoing projects and outputting time series data of risk scores for past and ongoing projects. The risk score calculation unit 20 uses the influence factor information 400, the encoded past project information 500, the division information lift value for each division information of each influence factor, and the ongoing project information as input information.

[1.1.4.1. Encoding of ongoing project information]
The risk score calculation unit 20 encodes the ongoing project information 800 to generate the encoded ongoing project information 800. The content of the encoding process is the same as the encoding of the past project information by the prediction model construction unit 10, and is stored in each storage field 811 to 815 of the ongoing project information 800 based on the classification information of the influence factor information 400. Replace the value that has been set with the corresponding partition value. However, in the ongoing project, the success / failure of the project is uncertain, so the project failure definition information 300 is not necessary here.

  FIG. 9 shows a data configuration example of the encoding in progress project information 900. The example shown in FIG. 9 is an example of the encoding ongoing project information 900 generated from the ongoing project information 800 shown in FIG.

  Similar to the ongoing project information 800, the encoding ongoing project information 900 includes five records 901 corresponding to the five ongoing projects NEW1 to NEW5. Each record 901 includes a “specification review indication density” storage field 911, a “design review indication density” storage field 912, a “design review efficiency” storage field 913, a “unit test bug density” storage field 914, and a “combination test”. It has a “bug density” storage field 915 and a “post-shipment bug density” storage field 916.

  The risk score calculation unit 20 stores the segment values corresponding to the values stored in the corresponding storage fields 811 to 815 of the ongoing project information 800 in the storage fields 911 to 915, respectively. However, if the value stored in the ongoing project information 800 is a value indicating “not implemented”, the value indicating “not implemented” is also stored in the corresponding storage field of the encoding ongoing project information 900. .

  In the example shown in FIG. 9, the labels of the storage fields 911 to 916 include the corresponding influence factor (variable) names “specification review indication density”, “design review indication density”, “design review efficiency”, “unit test bug”. X1, X2a, X2b, X3, X4, and Y are used instead of “density”, “joint test bug density”, and “post-shipment bug density”.

[1.1.4.2. Risk score calculation]
Based on the encoded past project information 900 and the ongoing project information 800, the risk score calculation unit 20 calculates a risk score using the timing information of each influential factor and the category information lift value for each category value.

The risk score calculation unit 20 calculates the risk score of the project p (regardless of whether it is in the past or in progress) as follows.
(1) The initial risk score S_p_0 of project p is calculated from the following formula 2.

In general, the initial risk score S_p_0 is set as described above assuming that there is no initial information regarding the project. However, if you want to change the initial risk score for each project, set the initial risk score individually. May be.
(2) The risk score for the timing information T = j of the project p is calculated.

  The risk score calculation unit 20 calculates the risk at T = j of the project p, where L_Xi_k is the classification information lift value where the value of the influence factor Xi of the project p is k, and T (Xi) is the timing information value of the influence factor Xi. The score S_p_j is calculated by the following formula 3.

However, if k = NA, L_Xi_k = 1. The second term on the right side of Equation 3 means the sum of the logarithms of the division information lift values for all influencing factors with timing information T = j. In addition, the value of k in Expression 3 is the partition value stored in the storage field corresponding to the influence variable Xi of the record 501 or the record 901 corresponding to the project p of the encoded past project information 500 or the encoded project information 900. It is.

  Assuming that ongoing projects show similar trends to past projects, it can be generally determined that the higher the project risk score, the higher the probability of project failure.

  Give an example of risk score calculation. A calculation example of the risk score of the project New1 included in the ongoing project information 800 and the encoding ongoing project information 900 shown in FIGS. 8 and 9 is shown.

  The risk score S_New1_1 of the project New1 for the timing information T = 1 is calculated as follows.

In addition, the value calculated by the prediction model construction unit 10 described above is used as the value of the division information lift value L_X1_0 in the above formula.
Further, the risk score S_New1_2 of the project New1 for the timing information T = 2 is calculated as follows.

In this example, since there are two influencing factors of timing information T = 2, X2a and X2b, the logarithm of the division information lift value of these two influencing factors is included in the calculation. The value calculated by the prediction model construction unit 10 is used as the value of the segment information lift value L_X2a_1 in the above formula. As the segment information lift value L_X2a_NA, a value = 1 when k = NA is used.

  The risk score S_New1_3 of the project New1 for the timing information T = 3 is calculated as follows.

なお、上記式中の区分情報リフト値L_X3_1の値は、前述の予測モデル構築部１０によって算出された値を使用する。

Note that the value calculated by the prediction model construction unit 10 described above is used as the value of the segment information lift value L_X3_1 in the above formula.

  The risk score S_New1_4 of the project New1 for the timing information T = 4 is calculated as follows.

  In addition, the value calculated by the prediction model construction unit 10 described above is used as the value of the segment information lift value L_X4_0 in the above formula.

  With the above, for the progress project New1, the risk score was obtained for each of timing information T = 1, T = 2, T = 3, and T = 4. As a result, it is possible to generate and output data indicating the time series change of the risk score in each timing information (T = 1,..., 4).

  FIG. 10 shows a line graph that visualizes the time-series change of the risk score in each timing information (T = 1, 2, 3, 4). From this graph, it can be seen that the predicted value of project failure risk gradually increases from T = 2 (design process) to T = 3 (unit test process).

  Similarly, risk scores can be calculated for all ongoing projects (New1-New5). By showing the risk score of each ongoing project for each timing information (T = 1, 2, 3, 4), the time series change of the risk score of each ongoing project can be visualized in a graph. FIG. 11 shows a graph showing the time series change of the risk score of the ongoing project (New1 to New5). From this graph, for example, the following information can be understood. Projects New3 and New4 have an increasing risk score, which is a predictive value of project failure risk. For these projects, it is desirable to strengthen monitoring. If necessary, it can be judged that risk reduction measures should be implemented for these projects.

  Project New5 can be considered to be relatively low risk at T = 2 (design process).

[1.2. Operation example of risk assessment device]
An operation example of the risk evaluation apparatus 1 shown in FIG. 1 will be described. FIG. 12 is a flowchart showing an example of risk score calculation processing which is the main operation of the risk evaluation apparatus 1.

  In the risk score calculation process, the risk evaluation apparatus 1, more specifically, the prediction model construction unit 10, generates encoded past project information 500 from the past project information 200, project failure definition information 300, and influence factor information 400 (S10).

  Next, the risk evaluation apparatus 1, more specifically, the prediction model construction unit 10 calculates a division information lift value for each division value for each influence factor based on the encoded past project information 500 and the influence factor information 400 (S 20).

  Next, the risk evaluation apparatus 1, more specifically, the risk score calculation unit 20 generates encoding ongoing project information 900 based on the pre-stored ongoing project information 800 and influence factor information 400 (S 30).

  Next, the risk evaluation device 1, more specifically, the risk score calculation unit 20, calculates the timing information of each influential factor and the category information lift value for each category information for the encoded past project information 900 or the ongoing project information 800. The risk score for each value of timing information is calculated for each project (S40).

  When the risk score output by the risk evaluation device 1 based on the past project information 200 shown in FIG. Similarly, for each of the other projects, the risk evaluation apparatus 1 outputs risk scores S_p_0, S_p_1, S_p_2,..., S_p_m (m is the maximum value taken by the timing information T) for each timing information value.

The risk evaluation apparatus 1 may output the risk score obtained as a result of the calculation in a numerical enumeration format or may output it as a graph. There is no limitation on the output format.
Above, description of the operation example of the risk evaluation apparatus 1 is complete | finished.

[1.3. Advantages of First Embodiment]
(1) Based on past project information composed of various measurement data, the risk of project failure in an ongoing project is quantified as a “risk score” that combines multiple measurement data, and the time series change is obtained. Can do.

  (2) By monitoring the time series change of the “risk score” representing the risk of project failure, it is possible to detect signs of project failure at an early stage and implement countermeasures at an appropriate timing.

[2. Second Embodiment]
A second embodiment of the present invention will be described. The second embodiment is proposed as a risk evaluation apparatus.

  In addition to the features of the first embodiment, the risk evaluation apparatus according to the second embodiment calculates the probability of project failure at an arbitrary point of time by logistic regression analysis using a risk score as an input. Features.

  The risk evaluation apparatus according to the second embodiment is an information processing apparatus such as a computer or a workstation. The information processing apparatus includes an arithmetic processing unit (CPU), a main memory (RAM), and a read-only memory (ROM). ), An input / output device (I / O), and, if necessary, an external storage device such as a hard disk device.

[2.1. Configuration example of risk assessment device]
FIG. 13 is a functional block diagram showing a configuration example of the risk evaluation apparatus according to the second embodiment.

  The risk evaluation device 2 includes an input information storage unit 30, a prediction model construction unit 10 connected to the input information storage unit 30, a risk score calculation unit 20 connected to the prediction model construction unit 10, and the risk score An ongoing project information storage unit 40 connected to the calculation unit 20 and a project failure probability calculation unit 50 connected to the risk score calculation unit 20 are included. The project failure probability calculation unit 50 corresponds to a third processing unit.

  The risk evaluation device 2 has basically the same components as the risk evaluation device 1 according to the first embodiment, and differs in that it further includes a project failure probability calculation unit 50.

  Among the constituent elements of the risk evaluation apparatus 2, the same constituent elements as those of the risk evaluation apparatus 1 according to the first embodiment are given the same reference numerals, and detailed descriptions thereof are omitted.

[2.1.1. Project failure probability calculator]
The project failure probability calculation unit 50 derives a failure probability regression equation based on the past project based on the time-sequential data of the risk score output from the risk score calculation unit 20 and the encoded past project information 500. It has a function to calculate the failure probability for an ongoing project using a stochastic regression equation and output the project failure probability for each timing information (T = 1, 2, ..., m).

  As for the time series data of risk scores, both past projects and ongoing projects are used. Further, the only information necessary for the encoded past project information 500 is a value of Y (project success / failure).

[2.1.1.1. Derivation of failure probability regression formula based on past projects]
Derivation of the failure probability regression formula based on the past project by the project failure probability calculation unit 50 will be described.

  The project failure probability calculation unit 50 stores a value (hereinafter, “success / failure determination value”) of a storage field (storage field 516 in the example shown in FIG. 5) that stores a value indicating success / failure of the encoded past project information 500. ) And the regression formula of the project failure probability based on the risk score of the past project for each timing information (T = 1, 2,..., M).

  Specifically, the project failure probability calculation unit 50 performs logistic regression analysis using the encoded success / failure determination value of the past project information 500 and the past project risk score in the timing information T = j as input. Find a and b in Equation 4. Logistic regression analysis may be implemented using standard statistical tools such as MINITAB and R.

In the above equation 4, “Pj” means the probability of success / failure determination value (Y) = 1 (failure) in the timing information T = j, and “X” means the risk score S_Z_j of the past project Z.

  Depending on the original data, it may not be possible to derive a formula by logistic regression analysis. In that case, the failure probability with the corresponding timing information is not obtained. For example, when the original data can be completely separated (success / failure can be completely identified with a certain threshold).

  Similar calculations are performed for other timing information (other than T = j).

  An example of derivation of a failure probability regression equation is shown. FIG. 14 shows original data for deriving a project failure probability regression equation by logistic regression analysis in the case of timing information T = 4. The original data 1400 includes success / failure determination values (values stored in the storage field 516) of each past project and a risk score S_p_4 of the corresponding past project p.

  The project failure probability calculation unit 50 obtains the results of a = 3.544, b = 7.342 from the original data 1400, and obtains the following failure probability regression equation.

The project failure probability obtained by Equation 5 is applied when the timing information T = 4. The project failure probability calculation unit 50 obtains a failure probability regression formula for each of other timing information.

[2.1.1.1. Calculation of project failure probability using failure probability regression formula]
The project failure probability calculation unit 50 calculates the project failure probability for the ongoing project using the obtained project failure probability regression equation. If the risk score of the target project is input using the failure probability regression equation of the timing information to be obtained, the project failure probability in the project can be obtained.

  An example of calculating the project failure probability is shown. In order to calculate the project failure probability P_New1_4 when the timing information T = 4 for the ongoing project New1 in the example shown in FIG. 8, the following calculation using the above equation 5 is performed.

In the above formula, X = −0.4121 (risk score of timing information T = 4 of ongoing project New1; calculated by the risk score calculation unit 20) is used.

  Similarly, the project failure probability calculation unit 50 can calculate the project failure probability in any timing information (T = 1,..., 4) for any ongoing project New1, New2, New3, New4, New5. is there.

[2.2. Example of operation of risk assessment device 2]
An operation example of the risk evaluation apparatus 2 will be described. FIG. 15 is a flowchart showing an example of a project failure probability calculation process which is the main operation of the risk evaluation apparatus 2 according to the second embodiment.

  In the project failure probability calculation process, the risk evaluation device 2 executes a risk score calculation process (S110). Since the risk score calculation process is the same process as the risk score calculation process (see FIG. 12, steps S10 to S40) in the first embodiment, the details of the process will be omitted.

  Next, the risk evaluation device 2, more specifically, the project failure probability calculation unit 50, derives a failure probability regression equation based on the past project information 200 (S120).

  Next, the risk evaluation device 2, more specifically, the project failure probability calculation unit 50 calculates the project failure probability of the ongoing project using the failure probability regression equation derived in step S120 (S130).

Finally, the risk evaluation device 2, more specifically, the project failure probability calculation unit 50 outputs the project failure probability calculated in step S130.
Thus, the risk evaluation device 2 ends the project failure probability calculation process.

[2.3. Advantages of Second Embodiment]
(1) By assuming that ongoing projects show similar trends as past projects, logistic regression analysis can be used to calculate the probability of project failure at any point in time for ongoing projects.

(2) The probability of project failure is easy to understand intuitively and can be used effectively for decision making for project control.
[3. Third Embodiment]
A third embodiment of the present invention will be described. The third embodiment is proposed as a risk evaluation apparatus.

  In addition to the features of the first embodiment, the risk evaluation device according to the third embodiment uses the expected category information lift value to select influential factors, set category information, etc. Is generated.

  The risk evaluation apparatus according to the third embodiment is an information processing apparatus such as a computer or a workstation. The information processing apparatus includes an arithmetic processing unit (CPU), a main memory (RAM), and a read-only memory (ROM). ), An input / output device (I / O), and, if necessary, an external storage device such as a hard disk device.

[3.1. Configuration example of risk assessment device]
FIG. 16 is a functional block diagram showing a configuration example of the risk evaluation apparatus according to the third embodiment. The risk evaluation device 3 includes an influence factor candidate information storage unit 60, an influence factor information determination unit 70 connected to the influence factor candidate information storage unit 60, and an input information storage unit 30 connected to the influence factor information determination unit 70. The prediction model construction unit 10 connected to the input information storage unit 30, the risk score calculation unit 20 connected to the prediction model construction unit 10, and the ongoing project information storage connected to the risk score calculation unit 20 Part 40. The influence factor candidate information storage unit 60 corresponds to a third storage unit, and the influence factor information determination unit 70 corresponds to a fourth process.

  The risk evaluation device 3 has basically the same components as the risk evaluation device 1 according to the first embodiment, and further includes an influence factor candidate information storage unit 60 and an influence factor information determination unit 70. Is different.

  Among the constituent elements of the risk evaluation apparatus 3 according to the third embodiment, the same constituent elements as those of the risk evaluation apparatus 1 according to the first embodiment are denoted by the same reference numerals and are the constituent elements. The detailed description of is omitted.

[3.1.1. Influencing factor candidate information storage unit]
The influence factor candidate information storage unit 60 has a function of storing the influence factor candidate information 1200. The influence factor candidate information 1200 is a set of candidates for influence factors and their classification information. Based on the influence factor candidate information 1200, which is information in which an operator, a user, an administrator, and the like of the risk evaluation device 3 input a plurality of influence factor and its classification information candidates, the influence factor candidate information storage unit 60 includes each influence factor and An expected category information lift value is calculated for the category information, an influence factor and its category information candidate are compared based on the expected category information lift value, and an influence factor to be included in the influence factor information 400 and its category information are determined. The timing information is not required for determining the influence factor and its classification information, but may be included in the influence factor candidate information 1200 in advance.

[3.1.2. Influencing factor information determination unit]
The influence factor information determination unit 70 encodes past project information, calculates an expected category information lift value, selects an influence factor to be included in the influence factor information 400 and its category information based on the expected category information lift value, It has a function of generating influence factor information 400 including the selected influence factor and its classification information and storing it in the input information storage unit 30.

[3.1.2.1. Encoding past project information]
The influencing factor information determination unit 70 encodes the past project information 200 and generates encoded past project information. A method for encoding the past project information 200 is the same as that in the first embodiment. However, the classification information used in encoding uses the influence factor candidate information 1200.

[3.1.2.2. Calculation of expected category information lift value]
The influencing factor information determination unit 70 uses the following expression 6 for each influencing factor information candidate (influencing factor Xi, classification information k = {0, 1,..., N}) included in the influencing factor candidate information 1200. Calculate the defined expected category information lift value EL_Xi.

Where in Equation 6

  The influence factor information determination unit 70 refers to the calculated expected category information lift value EL_Xi, determines which influence factors included in the influence factor candidate information 1200 are included in the influence factor information 400, and determines The influential factor information 400 including the influential factor is generated and stored in the input information storage unit 30.

  If there are multiple candidates for influencing factor information, it is considered that selecting a higher expected category information lift value EL_Xi is likely to increase the accuracy of project success / failure prediction. The present embodiment is not limited to the method of selecting the information lift value EL_Xi in descending order.

  A specific example of calculating the expected category information lift value is shown. Expected classification information lift value EL_Xi is calculated and compared for two candidate influence factors having different classification information for the same influence factor Xi.

(1) 1st candidate The 1st candidate has the contents of influence factor X1, division information {0: [MIN, 2.5), 1: [2.5, MAX], NA: missing value}.

  The expected classification information lift value EL_Xi of the first candidate is calculated. However, it is assumed that the past project information 200 uses the data shown in FIG.

  The influence factor information determination unit 70 generates the encoded past project information and the segment information lift value from the past project information 200 and the first candidate segment information, and the following values are calculated from the encoded past project information and the segment information lift value. Get.

  The influencing factor information determination unit 70 applies these values to the above equation 6 to obtain the first candidate expected category information lift value EL_Xi as follows.

(2) Second Candidate The second candidate has contents such as influencing factor X1, classification information {0: [MIN, 2.0), 1: [2.0, MAX], NA: missing value}. Although it is the same influence factor as a 1st candidate, the content of division information differs.

  The expected classification information lift value EL_Xi of the second candidate is calculated. However, as in the case of the first candidate, the past project information 200 is assumed to use the data shown in FIG.

  The influence factor information determination unit 70 generates the encoded past project information and the segment information lift value from the past project information 200 and the second candidate segment information, and the following values are calculated from the encoded past project information and the segment information lift value. Get.

  The influencing factor information determination unit 70 applies these values to the above equation 6 to obtain the second candidate expected category information lift value EL_Xi as follows.

When the expected category information lift value EL_Xi of the first and second influence factor information candidates is compared,
The second candidate and the expected classification information lift value EL_Xi (= 1.4441) are slightly higher than the expected classification information lift value EL_Xi (= 1.3443) of the second candidate. Therefore, in determining the influence factor information 400, the influence factor information determination unit 70 adopts the second candidate and includes the data of the second candidate in the influence factor information 400. This may increase the accuracy of project success / failure prediction.

[3.2. Example of operation of the third embodiment]
An example of operation of the risk evaluation apparatus 3 shown in FIG. 16 will be described. FIG. 17 is a flowchart showing an example of the main operation of the risk evaluation apparatus 3 according to the third embodiment.

  First, the risk evaluation apparatus 3, more specifically, the influence factor information determination unit 70 generates encoded past project information based on the past project information 200, the project failure definition information 300, and the influence factor candidate information 1200 (S210).

  Next, the risk evaluation device 3, more specifically, the influence factor information determination unit 70, expects the expected category information lift value for each of the influence factor candidates included in the influence factor candidate information 1200 from the encoded past project information generated in step S 210. EL_Xi is calculated (S220).

  Next, the risk evaluation device 3, more specifically, the influence factor information determination unit 70 determines the influence factor and its classification information to be included in the influence factor information 400 based on the expected classification information lift value EL_Xi, and the determined influence factor and its classification New influence factor information 400 including the information is generated and stored in the input information storage unit 30 (S230).

Next, the risk evaluation device 3 executes a risk score calculation process (S240). Since the risk score calculation process is the same process as the risk score calculation process (see FIG. 12, steps S10 to S40) in the first embodiment, detailed description of the process content is omitted. The influence factor information 400 used in the risk score calculation process (S240) is the influence factor information 400 generated and stored in step S230.
This is the end of the description of the operation of the risk evaluation apparatus according to the third embodiment.

[3.3. Advantages of the third embodiment]
In the construction of the prediction model of the first embodiment, work such as selection of influence factors and setting of classification information that is performed by a user or an operator takes the most labor, and the good or bad is the final project failure prediction accuracy. Also greatly affects.

  In the third embodiment, it is possible to select an optimal influence factor and set the category information using the expected category information lift value as an evaluation function, thereby greatly reducing the labor of the prediction model construction work. In addition to reducing this, it can be expected to improve the accuracy of project failure prediction.

[4. Fourth Embodiment]
Even if the first, second, and third embodiments are combined, the risk evaluation apparatus is established. FIG. 18 is a functional block diagram showing an example of the configuration of the risk evaluation apparatus 4 in which the first, second, and third embodiments are combined. Note that the same reference numerals are given to the same components as those in the first, second, and third embodiments.

[5. Summary, etc.]
As mentioned above, although embodiment of this invention was described, this invention is not limited to these, A various change, addition, a combination, etc. are possible in the range which does not deviate from the meaning of invention.

1, 2, 3, 4, risk assessment device, 10, prediction model construction unit, 20, risk score calculation unit, 30, input information storage unit, 40, ongoing project information storage 50 ... project failure probability calculation unit; 60 ... influence factor candidate information storage unit; 70 ... influence factor information determination unit

Claims (6)

Past project information, which is information describing the values of multiple influencing factors for projects implemented in the past,
Project failure definition information, which describes the criteria for distinguishing projects that were implemented in the past from successful projects and failed projects,
A first storage unit that stores classification factor information obtained by dividing an area in which a value of an influence factor can be divided into a plurality of divisions, and influence factor information that is information including timing information that indicates when the value of the influence factor can be acquired; 1 storage means;
Based on the division information of the project failure definition information and the influence factor information, the past project information is generated and replaced with the division value corresponding to the division to which the value of each influence factor belongs. A first processing means for calculating, for each category value of each influence factor, a category information lift value that is information indicating the degree of influence of the influence factor on the success or failure of the project, based on past project information;
Second storage means for storing in-progress project information, which is information describing a plurality of influencing factor values for the in-progress project;
Based on the category information of the influential factor information, by replacing with the segment value corresponding to the category to which the value of each influential factor of the ongoing project information belongs, the encoded in-progress project information is generated, and the encoded past Based on the project information and ongoing project information, the division information lift value, which is the information indicating the degree of influence of the influencing factors on the success or failure of the project, is calculated for each influencing factor timing information and the dividing value. Second process of calculating risk score for each timing information value for each project using timing information of each influencing factor and category information lift value for each category information for project information or ongoing project information And a risk evaluation apparatus. The second processing means calculates a division information lift value L_Xi_k having an influence variable of Xi and a division value of k by the following equation:

(However, c1 is the number of project records in which the value of the storage field indicating success or failure of the project is a value indicating success for the influence factor Xi,
c2 is the number of project records whose impact factor Xi has a partition value other than k and other than “NA”, and the value of the storage field indicating the success or failure of the project is a value indicating success,
c3 is the number of project records in which the partition value for the influencing factor Xi is k, and the value of the storage field indicating success or failure of the project is a value indicating failure,
c4 is the number of project records whose impact factor Xi is other than k and NA, and the value of the storage field indicating the success or failure of the project is a value indicating failure,
a and b are arbitrary numbers)
The second processing means calculates the risk score using the timing information of each influencing factor and the category information lift value for each category value according to the following formula:
The risk evaluation device according to claim 1

(However, S_p_0 is the risk score when the value of timing information of project p is 0,
S_p_j is the risk score when the timing information value of project p is 0, T (Xi) is the value of timing information of influence factor Xi, L_Xi_k is the division information lift value with influence factor Xi and division value k Is).   Based on the success / failure judgment value that is a value indicating the success or failure of the project in the encoded past project information and the risk score of the past project for each value of the timing information, the regression formula of the project failure probability is obtained, and the obtained project failure probability regression formula The risk evaluation apparatus according to claim 1, further comprising: a third processing unit that calculates a project failure probability for an ongoing project using. The third processing unit performs logistic regression analysis using the encoded past project information success / failure judgment value and the past project risk score in the timing information T = j as input, and a and b in the following equations The risk evaluation device according to claim 3, wherein a risk probability regression equation is obtained to obtain

(However, Pj is the probability of project failure in the timing information T = j). Third storage means for storing influential factor candidate information, which is a set of influential factor and its classification information candidates;
Generate encoded past project information from past project information using the influence factor candidate information, and expect the reciprocal of the classification information lift value when the classification information lift value or the classification information lift value is 1 or less Calculating an expected category information lift value that is a sum of the values, selecting an influence factor and its category information to be included in the influence factor information based on the expected category information lift value, and selecting the selected influence factor and its category The risk evaluation apparatus according to claim 1, further comprising a fourth processing unit that generates influence factor information including information. 6. The risk evaluation apparatus according to claim 5, wherein the fourth processing means calculates an expected category information lift value by the following equation.

Images