Forecasting Software Modules with Known Defects through the Quadratic Discriminant Analysis Feature Reduction Technique

Software validation is crucial in software advancement to ensure defect-free software. Software Defect Prediction aims to identify defect-prone modules at an early stage. Machine learning techniques with SDP have recently gained traction, offering the potential to reduce maintenance costs. Feature reduction can improve the efficiency of an SDP model. Selecting a suitable FR method can be challenging. This paper proposes a novel FR technique based on Quadratic Discriminant Analysis (QDAFR) to achieve the best features. The proposed technique was implemented on four supervised learning classifiers: Naive Bayes (NB), Decision Tree (DT), Support Vector Machine (SVM), and K-Nearest Neighbor (KNN) on six public NASA datasets. The conduct of the recommended technique was correlated with extant FR techniques: Principal Component Analysis (PCAFR), Linear Discriminant Analysis (LDAFR), and Kernel Principal Component Analysis (KPCAFR). Our experimental findings exhibited that the proposed technique gave higher accuracy and lower error rates with SVM.