A MACHINE LEARNING-BASED FRAMEWORK FOR IMPROVED SOFTWARE DEFECT PREDICTION

Abstract

Defect prediction is one of the active areas in the software engineering field. Closing the gap between data mining and software engineering is crucial to the product's success. Prior to testing, software defects prediction predicts source code faults. The impact area in software is often explored using techniques for predicting software flaws, including clustering, statistical approaches, mixed algorithms, metrics based on neural networks, black box testing, white box testing, and machine learning. This study's primary contribution is the first use of feature selection to improve machine learning classifiers' performance in predicting faults. This project aims to increase the accuracy of defect prediction in five NASA data sets: CM1, JM1, KC2, KC1, and PC1. The public may access these NASA data sets. In order to achieve a higher defect prediction accuracy than without feature selection (WOFS), this study combines the feature selection technique with machine-learning techniques, including Random Forest, Logistic Regression, Multilayer Perceptron, Bayesian Net, Rule ZeroR, J48, Lazy IBK, Support Vector Machine, Neural Networks, and Decision Stump. The aforementioned classifiers are used, da-ta is refined, and data is preprocessed using the research workbench, a machine-learning program named WEKA (Waikato Environment for Knowledge Analysis). A mini tab statistical tool is used to evaluate statistical studies. The study's findings show that, when compared to WOFS, the accuracy of defects prediction using feature selection (WFS) is improved.