A Package Based Clustering for enhancing software defect prediction accuracy

Software defect prediction models considering clustering are to combine related features to enhance the probability of predicting defects. Aggregating related and similar classes is the main challenge in software clustering. An efficient clustering approach named as Package Based Clustering has been proposed to group the software for predicting defects. It uses Object Oriented classes´ relationships and similarities to group the software into multiple clusters. To segregate a software project into multiple clusters, it performs textual analysis to identify all Object Oriented classes from the software project. Then it uses package information of each class to divide those into clusters. To analyze the proposed clustering algorithm, the linear regression model is used which learns from clusters of related and similar classes. The experiment has been conducted on JEdit 3.2 and results show that the prediction model using Package Based Clustering is 54%, 71%, 90% better than the prediction models built on BorderFlow clustering, k-means clustering and the entire system respectively.