A novel approach for software defect prediction using metaheuristic models

Abstract

Software defect prediction is the process of identifying defects in software modules to enable faster and more efficient testing process. Prior identification of prospective defective modules leads to better analysis of the identified modules. Testing can be concentrated on these defective modules, hence providing better focus on the defective modules. Such indications can also be highly useful for guiding the code review process, hence providing better quality assurance for the software being analysed.Increasing complexity of software systems has led to the increased necessity for analysing and testing software for their efficient functioning. This thesis presents four contributions to enable effective software defect prediction. The initial works concentrate on using SVM for prediction and metaheuristic models for hyperparameter fine-tuning, and the final work concentrates on developing a metaheuristic model using Hybrid Firefly algorithm for prediction. The initial contribution uses SVM for prediction. However, SVM is highly sensitive to parameters. Hence effectively fine-tuning the parameters can lead to more enhanced results. This contribution uses hybridized GA to fine-tune the hyperparameters of SVM. Hybridization in GA has been performed with BFO algorithm. The model exhibited improved predictions, however, it was found to be computationally complex. The second contribution aims to reduce the complexity levels by utilizing Cuckoo Search as the hyperparameter fine-tuning algorithm. The Cuckoo Search model was also hybridized using BFO to eliminate the issue of local optima. This model was observed to exhibit enhanced predictions compared to the previous model. newline