Analysis of Developers’ Network and Change Burst Metrics as Predictors of Software Faults

Abstract

Introduction: Many software quality metrics that can be used as proxies of measuring software quality by predicting software faults have previously been proposed. However determining a superior predictor of software faults given a set of metrics is difficult since prediction performances of the proposed metrics have been evaluated in non–uniform experimental contexts. There is need for software metrics that can guarantee consistent superior fault prediction performances across different contexts. Such software metrics would enable software developers and users to establish software quality. Objectives: This research sought to determine a predictor for software faults that requires least effort to detect software faults and has least cost of misclassifying software components as faulty or not given developers’ network metrics and change burst metrics. Methods: Experimental data for this study was derived from Jmeter, Gedit, POI and Gimp open source software projects. Logistic regression was used to predict faultiness of a file while linear regression was used to predict number of faults per file. Results: Change burst metrics model exhibited the highest fault detection probabilities with least cost of mis-classification of components as compared to the developers’ network model. Conclusion: The study found that change burst metrics could effectively predict software faults.