A package-based clustering approach to enhance the accuracy and performance of software defect prediction

摘要

To enhance the accuracy and performance, software defect prediction models considering clustering of dataset combine related and similar features to improve the learning process of the model. Here, a clustering approach named package-based clustering has been proposed to group the similar and related parts of software using object oriented classes' relationships and similarities. To segregate software into clusters, it performs textual analysis to identify all object-oriented classes from source codes. Then it uses package information of each class to divide those into clusters. To analyse the performance of the proposed algorithm, linear regression model is used, which learns from clusters of related and similar classes. The experiment has been conducted on eight releases of two open source software, which are Xalan and Ant, and results show that the proposed technique outperforms the existing clustering algorithms those are BorderFlow and the entire system.