A Novel UML Based Approach for Early Detection of Change Prone Classes

This article describes how predicting change-prone classes is essential for effective development of software. Evaluating changes from one release of software to the next can enhance software quality. This article proposes an efficient novel-based approach for predicting changes early in the object-oriented software. Earlier researchers have calculated change prone classes using static characteristics such as source line of code e.g. added, deleted and modified. This research work proposes to use dynamic metrics such as execution duration, run time information, regularity, class dependency and popularity for predicting change prone classes. Execution duration and run time information are evaluated directly from the software. Class dependency is obtained from UML2.0 class and sequence diagrams. Regularity and popularity is acquired from frequent item set mining algorithms and an ABC algorithm. For classifying the class as change-prone or non-change-prone class an Interactive Dichotomizer version 3 ID3 algorithm is used. Further validation of the results is done using two open source software, OpenClinic and OpenHospital.