COMPARATIVE ANALYSIS OF CLASSIFICATION ALGORITHMS FOR DEFECT PREDICTION

Abstract

The objective of this thesis is to evaluate the performance of the classification algorithms on binary classification problems using a variety of performance metrics: classification accuracy, precision, recall (sensitivity), F-Measure and ROC area. The evaluation is performed with an intention to identify which algorithm suits best for prediction of defect prone classes in software based on software quality metrics. Motivation As independent testing team, it is important to plan and manage the test execution activities in order to meet the tight deadline for releasing the software to end-users. Since the aim of test execution is to discover as many defects as possible, testing team is usually put into burden to ensure all defects are found and fixed by the developers within the system testing phase. Additional number of days has to be added to the timeline to accommodate testing team in completing their test with the hope that all defects have been found and fixed. On the other hand, the stakeholders would also ask the testing team on the forecasted defects in the software so that they could decide whether the software is feasible and fit for release. This is due to the nature that system testing is the last gate before the software is made visible to end-users, thus as the custodian of executing system testing, the independent testing team has to take responsibility to ensure software to be released is of high quality.