Multi-criterion Intelligent Decision Support system for COVID-19

3.1. Decision making

In an optimal engineering design environment as such, solving the multi-criterion decision-making (MCDM) [8], [9] problem is considered as a combined task of optimization and decision-making process to fight from the disease in a manner to stop spreading its effect on a large number of people.

Before, making final decision some adequate actions needs to be performed to choose the best out of so many alternatives present for solving the problem some criterion’s needs to be looked upon. This criterion in an organization comes from history, norms objectives, and first step. It all comes from data mining. So, to apply this theory, we have to set the following components to make it happen, i.e., we need:-

• •

  Decision-makers/Deciders

• •

  Criterion comparison

• •

  Alternatives (possible solution to our problems)/ classify options based on criterion

There could be number of deciders in the group like: - input deciders, normal deciders, and the important deciders in the group. The action to be taken for the particular group or society as the need or severity of the disease can be different in different areas [10] on how to save the mankind from this on-going pandemic situation.

Criterion comparison of every decider must know which one is the most important to take the rest of the decision. And; finally, we need to evaluate every decision to be an alternative based on every criterion. Criterion impacts the selection of other options. Criterion [11] include the following:

• •

  Completeness – All the types of parameters should be included in it. It acts as a closure property of mathematics.

• •

  Redundancy – All the duplicates should have been removed at early stages only so as redundant data can be avoided efficiently.

• •

  Operational – Each alternative should be compared against every option in decision-making criterion.

4.1. Discussions

The amount of data produced and communicated over the Internet is significantly increasing, thereby creating challenges for the organizations that would like to reap the benefits of analyzing this massive influx of big data. It is crucial to examine [12] which data sets can be converted from raw data to complete information that may be proved ubiquitous for humankind to grow fast and flatten the growing curve of cases among people.

• •

  Loading info Establish the whole right information set to the well execute of the method. Load all the information into the data sets and implement the method well.

• •

  Calculate results use above loaded data to calculate the results and to get the best solution from the problem.

Refer Fig. 2, in the methodology section, when there is a massive influx of raw data and we are interested in calculating the best results. So, the loading info, loads all the data in raw form or unstructured form then exploration of data takes place by implementing various methods and patterns to it. Then, that raw data is transformed from unstructured form to structured form and then the results are interpreted.

The Table 2 shows various methods with their advantages, disadvantages and their implications in different areas. Every method has its own advantages and disadvantages. Different methods are used in different fields to produce results in real simulations for taking the decision taking in cause various methods as in Simple Additive Weighting (SAW) method the main area of application is financial management where this method is used. Technique for Order Preference by Similarity to the Ideal Solution (TOPSIS) method is generally used in areas where programming language is used to take decision so this method is majorly used in areas of engineering and marketing where diverse nature needs to be looked upon. Analytical Hierarchy Process (AHP) method is easy to use and data is not focussed much. This is entirely used in planning related tasks. Simple Multi Attribute Rating Technique (SMART) is simple technique where the decision maker’s makes less effort entirely used in transportation and logistics related problems. The fuzzy set theory is used in areas where the input is the major information to be analyzed upon and fuzzy sets are basically used in the fields of electronic health related data sets. The Elimination at Choice Translating Reality (ELECTRE) considers all vagueness and uncertain data and makes it more certain and efficient to be used in energy and transportation related problems.

4.2. Steps to manage and process data

Big Data represents a new era in data exploration and utilization. While most data is stored in hard copy form, the current trend is toward the rapid digitization [13] of these massive amounts of data.

• •

  Establish the number of deciders and alternatives.

• •

  Establishing criterion

• •

  Deciders choose their weights by consensus.

• •

  Every decider differentiates every settled criterion with the others.

• •

  Getting the consistent weights of criterion’s.

• •

  Every decider evaluates every alternative based on the criterion.

• •

  Weighting of the alternatives with the consistent weights of each criterion.

• •

  Group aggregations.

• •

  Weighting of the resulting values on step 7 with consistent weights of deciders.

• •

  Establishing max & min values.

• •

  Distance between the final values with the max & min value.

• •

  Calculate the final coefficient of the alternatives.

Firstly, start from formulating options, and decision-makers [11] must identify the decision goal, record the observations of all criterion’s in a table by analysis process. The decision maker’s then brainstorming, considering all the requirements to attain the best decision or the solution for the current problem. Then, it is to determine the decision alternatives, followed by the criterion. The decision set-up is entirely based upon the numerical values and also takes into account the representation of all the processes. The step before final decision to be taken depends upon the decision structuring process. Then, the final decision is taken. After; all the criterion weights or scores are computed, the decision-makers pairwise compare every alternative with respect to each of the criterion. Quantitative and qualitative measures are appropriately taken for each sets of criterion using the comparative approach and the numerical analysis of each value. Each analysis tool follows the different methods to calculate the final results of the multi-criterion decision-making approach. Many different techniques have emerged, which yields different results applied to an identical problem and helps in taking mitigate measures before the situation worsens (see Fig. 3).

4.3. Components of multi-criterion decision analysis

Multi-criterion decision analysis is a valuable asset that we apply for making complex decisions as needed in the case of COVID-19. It is mostly responsible and used in solving problems that have a choice among alternatives. It is a useful decision support tool, i.e., logical, consistent, and easy to use.

Multi-criterion decision analysis problems are composed of five components:-

• •

  Goal

• •

  Decision-makers

• •

  Decision Alternative

• •

  Evaluation Criterion

• •

  Outcomes or results

4.4. Factors

The factors needed to be analyzed are:-

• •

  BCG (Bacillus Calmette Guerin) $=$ in some countries like India BCG vaccine is a necessary vaccine given in childhood to infants, which is used for treating TB (Tuberculosis). People vaccinated with BCG [15] are proved to be least prone to the vaccine as it provides more immunity [7] to the body and helps the body to fight viral infection by making antibodies around it; if it may enter. It is observed that countries having policies of BCG vaccination [15], [16] found a lower number of confirmed cases, and a lower death toll is observed. Although; data varies from country to country, and it also varies among different age groups. The vaccine founds to be ineffective in adulthood.

• •

  Nutritional value – it is often said that citrus fruits or the food having nutrients like vitamin C, vitamin D, and magnesium ions in it proves to be ubiquitously resourceful for providing immunity against virus. So, health ministers and doctors strongly encourage people to boost immunity against the virus so that the body prepares itself for the defense mechanism by making antibodies and fight the disease well. Moreover, results from the sensitivity analysis corroborate that people with adequate immunity are less prone to virus and are seen even to recover if caught in an infection at a mild stage only.

• •

  Ministry of AYUSH – COVID-19 is SARS (Severe Acute Respiratory Syndrome) viral infectious disease which causes – “Corona virus”. At mild stage, ministry of AYUSH says it could be easily controlled with just home-made remedies. Medicines [17], [18], [19], [20] made from home ingredients, natural herbs like Tulsi, Ashva gandha etc. helps in boosting immunity that helps to fight the virus. Yoga and breathing control exercises are found to be the effective measures in curing the viral disease accompanying Ayurveda, naturopathy, unani, siddha and homeopathy as well.

• •

  Recently, Glenmark pharmaceuticals [21] on June, 20 have received approval from the drug controller general of India to manufacture and market Favipiravir, an antiviral drug to treat Covid-19 patients. Favipiravir is an antiviral drug that is approved in Japan for treating influenza and is also under the processing in many clinical trials; some trials have proved positive results while still are under the human based trial.

4.5. Results

The above discussed methodology about multi criterion decision making (MCDM) is a process of evaluating real-world situations, based on various qualitative or quantitative criterion in uncertain & risky environment to suggest an action or strategy or choosing the best option out of many available options. So, now to understand about eight worst-hit states due to COVID-19 we will be taking in cause the MCDM approach (see Fig. 5).

The data is analyzed on the Weka tool to see the number of people deceased from the total population. The data of the top 8 states in which the effects of COVID-19 is seen on a large scale, the data for the total population is taken. Then out of the total population number of confirmed cases, cases recovered, and the number of people died, statistics have been  [3], [22] observed to see the percentage of people affected out of the total population in the particular state.

So, Table 3 is experimental on the Weka [11] tool showing results as out of a total population in all the eight states, 0.537% of people died, and rest are recovered even after being affected. Through the above depicted data, we conclude that India encountered least infectious mutant form because of which the recovery rate is quite high. Many people are seen to recover  [23] from the virus in its mild stage only. The reason could be the nutritional values Indians consume since birth. Also, the vaccine BCG (Bacillus Calmette Guerin) is one of the factor proven to one of the recovery measure against the virus. Infants from birth to ten weeks of age are injected with BCG vaccine [3] so that the measures of high fever or TB (Tuberculosis) can be prevented (see Fig. 6).

Then, the K-means cluster analysis [24] is also performed on the data sets showing the statistical analysis, i.e., mean and standard deviation analyzed in the training table. The above analysis works on the five attributes i.e., states, population, confirmed, recovered, and deaths. Evaluation model evaluates on all training data sets. The above analysis method uses “Best first” technique for search based method. It uses forward search direction. Attribute subset evaluator evaluates each attributes individually using supervised learning model. It uses CFS subset evaluator as the feature selection algorithm. After performing, CFS pre-processing the number of steps are significantly reduced and more precise build-in classifications are obtained. Data set is divided into attributes and evaluates attributes by the degree of consistency.

After, attribute selection statistical analysis [25] takes place where mean and standard deviation on all four parameters such as: - Total population, confirmed cases, recovered cases, and the number of people died due to the pandemic (see Table 7).

Here, using Weka  [26] we also analyzed mean and standard deviation of all four parameters:-

Using the above formulas, results are obtained (see Fig. 7).

Here, for analyzing clustered k-means analysis methods is chosen with 100% precision working on all eight instances of Table 4. We also calculate the correlation coefficient, which is a numerical measure of some types of correlation in the total population, confirmed cases encountered, recovered cases, and the number of people who died. Then, the mean absolute error, root mean absolute error, relative absolute error and total relative squared error is analyzed on the all right instances or the top eight states where the COVID-19 is spreading in a massive manner for public health emergency for fast-spreading and endangering a large number of health of the people, and it requires immediate actions to prevent the disease at the community level. The patterns of public sentiment on analyzing related health information and assess influenza at the political and economic spread of the virus.

We obtain clustered instances on k-means clustering method on all eight instances and the value of log likelihood suggests that the variables are discrete. This means that analysis obtained on the type of graph is discrete type variables. So, the value of log likelihood is −20.43 (approximately) showing discrete natured variables in the file (see Fig. 8).

As, we calculated statistical analysis following results are obtained on all the eight instances. This shows disease is a fast-spreading disease that is spreading at the community level, and immediate steps have to be taken to prevent the spreading [27], [28] of the disease at the community level. The disease is a health threat to humankind WHO [29] declared it a pandemic disease researchers from the different institutions are trying to make an Antidote for the disease so that large spread could be saved. The chart shows out of the total population number of confirmed cases found, cases recovered, and the number of people died linearly. We can see from the graph that linear regression in deaths has been shown, which even can be controlled if we take immediate measures for the people to solve the extreme situation (see Fig. 9).

The above table predicts the results of the fatality rate of people state wise. It takes into account the number of people died in a particular state due to pandemic till now as this number is increasing to a great extent. We can see that Maharashtra is the worst hit state still we can predict that the number of people died are very less in comparison to positive cases. So, we can say that we can recover from the virus with improved recovery rate with appropriate prevention before situation worsens (see Fig. 10).

We concluded from the above graph that, the confirmed cases of COVID-19 are high in few states. The rate of recovery is appreciable that proves the fact that Indians have developed immunity against virus. But, still we have few cases of fatality rate generally among those people who are either aged, or in infant age group, or they are already suffering from any ailment related to breathing (see Fig. 11).

We have also calculated complete accuracy by class, including the precision values and weighted average value. [26] The TP rate i.e., the true positives (are the instances classified as a given class) that shows the value 1.000 that means all the instances are examined properly. The FP rate i.e., rate of false positives (are the instances falsely classified as a given class).

Precision is the instance that are calculated as the truly positives of a class divided by the total instance of the class.

The Recall factor is the proportion of instance calculated as the given class divided by the TP rate in that class (or the actual total in that class).

F- Measure is a combined factor of precision and recall.

Table 5 presents results obtained on three parameters i.e., Precision, Recall and F- measure by comparing varied Learning methods. After analyzing the values on Weka we have calculated improved results on all three parameters i.e. Precision, Recall, and F-measure (as shown in Table 5).

We also tried to analyze another three parameters i.e. MCC, ROC, and PRC.

MCC (Matthews Correlation Coefficient) is generally regarded as a balanced measure. It is even used with classes that are of very different sizes.

ROC (Receiver Operating Characteristics) area measurement gives an idea of how the classifiers are performing in the real simulation of data.

PRC (Precision–Recall) area gives more information than the ROC plot area. PRC plot [29] is generally used when we evaluate binary classifiers on variable datasets. PRC is more informative than MCC and ROC. [26]

The values of MCC (Matthews Correlation Coefficient), ROC (Receiver Operating Characteristics) and PRC (Precision–Recall) are calculated number of times in different cases i.e. I, II and III to improve the efficiency. It is estimated from the above analysis (Table 6) that in India, we have the least mutant form of infection in which if proper care and treatment happen, we can save the lives of all around us. As per the MOHFW report [], it is observed that people having diseases like hypertension and diabetes are more prone to the virus. Thus, special attention should be given to all those suffering from the ill-effects of it.

Confusion matrix is used to describe the performance of a classification model on a set of data sets assigned to understand the true values. It also helps in improving the performance of an existing data models by evaluating the data set.

So, in the above confusion matrix [25] obtained results are interpreted as:-

a $=$ Tested negative (It means the target variables are functional)

b $=$ Tested positive (It means the target variables are non-functional)

‘a’ and ‘b’ are the letters assigned to class values.

So, using these models improves the efficiency of class variables to a greater extent.

We can say that few factors needed to be taken in the cause before coming to a decision; decision-makers in multi-criterion decision  [14], [30], [31] making progress pose numerous alternatives before predicting a person may/may not possess a disease all the factors must be analyzed.

5.1. Proposed methodology

Now, we propose the model that also takes into account the number of people those are died, making it another parameter fatality rate F (t). F (t) is the proportion of infected people who will die due to the pandemic. It is an important parameter to be taken in cause to understand the threshold amount of people able to tackle the effects of the virus and also to track the number of people not able to struggle with the virus. This parameter F (t) will also help us to understand which level or what kind of historical background people are not immune to the virus or the special care to be taken upon such people as an immediate measure.

So, this SEIFR model will help the society to make better decisions in accurate time, which might not complicate efforts and pin down the number of people dying due to its ill effects. Through this number, we will be able to bridge the gap and help humankind in a better way.

5.2. SIR

Susceptible–Infected–Removed model, developed by Ronald Ross1, William Hamer. It was first used by Kermack and McKendrick in 1927. It was used for variety of infectious diseases such as: - Measles, airborne diseases.

5.3. SEIR

Susceptible–Exposed–Infected–Recovered model, developed by J.L. Aron and I.B. Schwartz. It is used to predict the infected numbers. It was used for variety of diseases like HIV etc (see Fig. 12).

S $=$ Susceptible

E $=$ Exposed

I $=$ Infected

F $=$ Fatal

R $=$ Recovered

This SEIFR model is the novel approach where we will also take in count the fatality rate so that we can improve the rate by predicting the conditions before they go out of hand.

The complete flow of the proposed SEIFR model is as shown in Fig. 13. The existing SEIR algorithm has been modified so that it can be applied to the given healthcare domain i.e., to calculate infectious fatality rate to analyze the death rate and other features for fatality rate prediction.

Hence, the modifications to an existing SEIR are proposed.

Under this section, the experimental setup, input parameters and implementation of SEIFR are discussed.

5.4. Terminology

S (t) is the number of individuals who are not affected and are even not immune to the disease at time t. If such people come in contact with another person who is the carrier of disease, they may also become exposed.

E (t) is the number of individuals who are infected, i.e., exposed to the virus, but these individuals are asymptomatic, showing no sign or symptoms.

I (t) is the infected individuals who are currently infected and can transmit the disease to others in near time t. These are the infected individuals which show the sign of virus and also transmit the infection to healthy individuals.

F (t) is the number of individuals who may fall under the category of death. Still, if some special precautions at some specific time t will be taken, then we can even save those individuals via decreasing the rate of fatality and increasing the number of recovered ones.

R (t) are the individuals those have become immune to virus and are successfully recovered from the virus, and these persons might have also developed immunity against the virus.

$\beta =$ infected individuals due to contact per day.

$\gamma =$ time period of infectious disease.

$\frac{1}{\gamma }=$ average period of infection

$\mu =$ per capita death rate

$\mu$N $=$ per capita birth rate

N $=$ S (t) + E (t) + I (t) + R (t) $=$ total population size

Mathematical models are the important tools for analyzing the control and spread of infectious disease. The disease that have longer incubation period are divided into subsequent layers such as: - Susceptible, Exposed, Infected, Fatal and Recovered. Above equations, shows the way to calculate the modeling of data in SEIFR model taking initial parameters as $\beta ,\gamma ,\&\mu$ as defined above. The unique differentiation of the above parameters yields the results of SEIFR model using the epidemiology of data. The model is run with ordinary differential equations. SEIFR Model tries to predict how a disease spreads or the number of infected people or the duration of epidemic or the onset of symptoms and the recovered stage (see Fig. 14).

In mathematical modeling [] of data, SEIFR layers interact i.e., Susceptible, Exposed, Infected, Fatal, Recovered. Infection starts from the susceptible layer that shows the beginning of the infectious state. The infection may spread to others in the time period of incubation i.e., onset of infection till the seven or nine days from the beginning denoted by $1∕\gamma$. This layer is called exposed layer where the droplet infection spreads to other healthy individuals. Then, after two–three days of infection the symptoms arises in the body that becomes the state of onset of symptoms or the infected state. Symptoms such as fever, nausea, cough, breathing problem, sneezing etc. arises. If the person successfully, completes the incubation period of nine days after the onset of infection. Then he/she may enter into recovered state or the person may enter into fatal state. Fatal state is the end of life that shows the immunity of a person is not strong enough that does not went into recovered state. If person successfully completes, incubation period time then recovered state arises from where the person will recover completely. Then, enters into healthy state. The model shows that the recovery rate is good in India. As Indians are showing great recovery against virus and we can even heal ourselves from the pandemic situation only if we take some immediate measures. Protective measures can only help in fighting from the pandemic such as: - wearing masks, maintaining social distance, washing hands with soaps to kill germs, avoid touching face, stay home, avoiding gatherings, and many more to keep oneself safe and healthy. As COVID-19 has led the world into an unprecedented state because there is no such symptoms. There are no specific symptoms  [30], [31] exhibited by the patient. Every patient undergoes treatment according to the symptoms arises. Even some are recovering on their own having mild symptoms with home quarantine (see Fig. 15).

Scenic map view of worst eight hit states through which we predict the confirmed cases out of total population using predictive driver analysis with 95% confidence level. In the total population given we can predict the number of confirmed cases out of total population using prediction driver analysis. If the same situation continues the Cognos IBM software predicts the rate at which the positive or confirmed cases can be predicted from total population.

Predictive driver analysis predicts the cases which can be controlled at great extent if the person knows the proper preventive steps or the measures so that he/she may save from being trap against the virus (see Fig. 16).

Predictive driver analysis out of total samples is performed that shows the probability of number of confirmed or positive cases out of total population. Different predictive strengths are obtained on different parameters. Different parameters matching’s are analyzed to calculate the best possible predictive strength for analyzing which was chosen to be 79%.

5.5. Predictive strength obtained on different parameters are

• •

  State is a predictor of total samples with the predictive strength of 41%.

• •

  Number of negative cases is a predictor of total samples with the predictive strength of 51%.

• •

  Number of positive cases is a predictor of total samples with the predictive strength of 60%.

• •

  The combination of positive cases and negative cases is a predictor of total samples with the predictive strength of 71%.

• •

  The combination of negative cases and states is a predictor of total samples with the predictive strength of 75%.

• •

  The combination of states, positive cases and negative cases is a predictor of total samples with the predictive strength of 79%.

Table 8 shows number of people will be found positive or under confirmed cases out of Total population if analyzed using the same predictive strength of 79% (see Fig. 17).

Bubble graph analyzed using predictive driver analysis on IBM Cognos software yields results of top worst eight hit states with their predictive strengths taking in cause the total population, positive cases and negative cases in particular state. Based on these factors, predictive strength is calculated.

5.6. Exploratory data analysis (EDA)

To summarize main characteristics from our dataset we used several python libraries (see Fig. 18, Fig. 20, Fig. 21, Fig. 22).

Here, through this graph, we can estimate the amount of susceptible, exposed, infectious, and recovered among the population affected due to COVID-19 the graph (Fig. 19) depicts the number of people exposed in various ways as we can see some people show susceptible changes. Exposed changes are those changes when people show remarks which even sometimes do not show symptoms. So, the number of exposed cases are high. People showing some infectious remarks generally tend to be recovered, which indicates the number of people recovered signifying the reasonable recovery rate if appropriate measures are taken. The fatality rate to be calculated depends upon the total population size subtracting the recovered ones showing that if nutritional value and other factors like vaccination and other needs since birth to be taken in the cause. We also predict that the Fatality rate is very less, which says that if appropriate measures are taken, we can achieve excellent results in favour of humankind. Thus, this model will contribute a lot in helping the society to have a life free of the pandemic situation by taking appropriate measures at the correct time.