Abstract

A software bug that is moved to the next stage of the software development lifecycle (SDLC) costs ten times more to remove. This lowers the quality of the final software product and makes the job of the project managers more challenging. As a result, the software industry has mandated that high-quality software projects must be completed on schedule and within budget Support vector machine (SVM) and artificial neural network (ANN), two classification techniques with the prediction power to manage the intricate non-linear correlations between the software characteristics and the software fault, have been empirically compared. Artificial neural networks (ANNs) are suitable to construct defect prediction models because of their capacity to manage the intricate nonlinear interactions between the software metrics and the defect data. The feature selection techniques solve this issue. Two classifiers, ELM and KELM, which are based on wrapper and filter-based feature selection techniques, are used to build SDP models. The study aims to ascertain two things: (1) the efficacy of feature selection-based classification models in software defect prediction; and (2) whether or not the elimination of superfluous features significantly alters the performance of the SDP models