Optimizing Defect Removal Efficiency by Defect Prediction using Machine Learning

Abstract

In current world, complexity and volume of software applications are increasing exponentially. Applications are expected to perform without defects as critical real world transactions are being handled through software design and development. Quality of a software can be impacted by software defects and thus leading to unavoidable high cost and customer dissatisfaction. Preventing defects at early stages of development ensures high quality software. Different defect prevention and detection techniques are used to identify the defects before delivery. In the last decade, machine learning models as defect detection techniques have taken a lot of attention from researchers as this concept narrows down the volume of code under inspection. In this research work, six machine learning algorithms are implemented. The prediction results are based on PROMISE public datasets containing more than ten thousand records. Performances of these algorithms have been compared through Confusion Matrix and Area Under the Curve (AUC) of Receiver Characteristic Operator (ROC) which are the most informative indicators of predictive accuracy in software defect prediction. The result analysis shows MLP is the best fit model in both CM and AUC-ROC showing maximum accuracy.