Noise Filtering and Imbalance Class Distribution Removal for Optimizing Software Fault Prediction using Best Software Metrics Suite

Abstract

Software fault detection is by far the most prevalent field of predictive analysis in the software engineering paradigm and several research centers have launched new ventures in this area. Predicting faults in software components before they are delivered to the end-users is of key importance, as it can save time, effort, and inconvenience associated with identifying and addressing these issues at later stages. This paper presents a software defect prediction technique to alleviate some basic problems of existing frameworks for predicting software defects. This study aims to combine simple noise removal, imbalanced class distribution, and software metrics selection techniques for optimizing defect prediction in software. The technique was tested on ten software fault prediction datasets. The experimental results including recall, precision, F-measure, accuracy, and ROC-AUC values show that the proposed method enhances fault prediction performance and the results obtained are better than or close to several comparative models. This proves the validity of our model.