An empirical evaluation of classification algorithms for fault prediction in open source projects

Abstract

Creating software with high quality has become difficult these days with the fact that size and complexity of the developed software is high. Predicting the quality of software in early phases helps to reduce testing resources. Various statistical and machine learning techniques are used for prediction of the quality of the software. In this paper, six machine learning models have been used for software quality prediction on five open source software. Varieties of metrics have been evaluated for the software including C & K, Henderson & Sellers, McCabe etc. Results show that Random Forest and Bagging produce good results while Naïve Bayes is least preferable for prediction.

Keywords

• Metrics
• Fault prediction
• Receiver Operating Characteristics Analysis
• Machine learning
• Nimenyi test