A New Approach to Find Predictor of Software Fault Using Association Rule Mining

摘要

In this paper, we use a new method to find the best predictor of software fault using association rule mining. The method, first of all select all the association rules having confidence greater than 40% and support greater than 30% using Apriori algorithm. After that our aim is to select top ?n? association rules out of a pool of ?k? association rules based on heuristic analysis. The method ranks association rules giving weight to a larger set of parameters than used by standard methods. The role of correlation has been emphasized in this method which also tries to eliminate issues faced in incorporating correlation, support and confidence expressively into a single fitness function. A least square regression analysis has been done to establish the best rules in a set of ?good? rules and allows for pruning of misleading rules that are often suggested by standard algorithms like the Apriori method. Furthermore, we investigate which OO-metrics are related to each other by best rules. The metrics on the antecedent part make sure the occurrence of the consequent part metrics. So, those OO-metrics which are present in the rule at antecedent part in most of the rules can be used as best predictor in software fault. It is found that applying this method results in both accurate and comprehensible rule sets as well as best predictor of fault.