Improving Software Quality Classification with Random Projection

Improving the quality of software products is one of the principal objectives of software engineering. Software metrics are the key tool in software quality management. In this paper we propose to use naive Bayes and RIPPER for software quality classification and use random projection to improve the performance of classifiers. Feature extraction via random projection has attracted considerable attention in recent years. The approach has interesting theoretical underpinnings and offers computational advantages. Results on benchmark dataset MIS, using Accuracy and Recall as performance measures, indicate that random projection can improve the classification performance of all four learners we investigate: Naive Bayes, RIPPER, MLP and IB1. With the help of random projection, naive Bayes and RIPPER are better than MLP and IB1 in finding fault-high software modules, which can be sought as potentially highly faulty modules where most of our testing and maintenance effort should be focused