Generating Adaptive Partially Ordered Sequential Rules

NOTICE OF CONCERN: ACM has received evidence that casts doubt on the integrity of the peer review process for the ICIA 2016 Conference. As a result, ACM is issuing a Notice of Concern for all papers published and strongly suggests that the papers from this Conference not be cited in the literature until ACM's investigation has concluded and final decisions have been made regarding the integrity of the peer review process for this Conference.

Sequential rule mining is an important data mining issue which has numerous applications. They are profoundly used in predicting the behaviour of learners in Educational data, predicting the web traversal patterns, finding the consecutive connections between gene expressions of different patients in Bio Informatics, determining the purchase pattern of customers in shop etc. Mining for sequential rules common to multiple sequences has some drawbacks such as strict ordering between items, because of which several rules may represent the same situation, similar rules are rated very differently and, rules may be too specific and less likely to be useful, sometimes none of the rules would match the new sequence. Thus, a more broad type of sequential rules common to multiple sequences, such that items in the forerunner and in the resulting of every rule are unordered, is required. These are called partially ordered sequential rules. (POSR). Rule Growth Algorithm and T-Rule Growth algorithm are used for mining the POSR. Both Rule Growth and TRule Growth Algorithm gives rise to lesser number of rules with greater prediction accuracy compared to mining sequential rules common to multiple sequences. The proposed work focuses on making these partially ordered sequential rules adaptive to the changes that occur over course of time. Two approaches are used to bring out adaptive rules. The first approach uses rating as the key parameter to eliminate the weakest rules and strengthen the stronger rules. The second approach classifies the rules based on their rule scores into different categories of strength and uses fuzzy inference system to infer the incremented rule scores. The performance of Rule Rating algorithm (Incremental approach) seems to have a better execution time with respect to Recomputation (i.e Rule Growth / T-Rule Growth algorithm applied for the entire dataset).