CODE: A Moving-Window-Based Framework for Detecting Concept Drift in Software Defect Prediction

Concept drift (CD) refers to data distributions that may vary after a minimum stable period. CD negatively influences models’ performance of software defect prediction (SDP) trained on past datasets when applied the new datasets. Based previous studies SDP, it is confirmed accuracy models affected due changes in distributions. Moreover, cross-version (CV) are naturally asymmetric nature their class imbalance. In this paper, moving window-based concept-drift detection (CODE) framework proposed detect chronologically defective and investigate feasibility alleviating from data. The CODE consists four steps, which first pre-processes forms CV chronological data, second constructs models, third calculates test statistics, fourth provides hypothesis-test-based method. prior observed an effort make more symmetric, class-rebalancing techniques utilized, improves models. ability demonstrated by conducting experiments 36 versions 10 projects. Some key findings are: (1) Up 50% chronological-defect drift-prone while applying most popular classifiers used SDP literature. (2) had positive impact for CVDP correctly classifying modules detected up 31% resampled