A Framework of Software Defect Prediction using Machine Learning with Updated Firefly Algorithm and Neural Networks

Abstract

This research investigates the impact of using swarm intelligence algorithms for feature selection in software defect prediction models. Leveraging the JM1 software dataset, comprising 10,000 records, the study evaluates the performance of three swarm intelligence algorithms—Firefly Algorithm, Cuckoo Search, and Particle Swarm Optimization (PSO)—in combination with Deep Neural Networks (DNN). The study focuses on three critical metrics: Recall, Precision, and F-measure. Results indicate that the Firefly + DNN model achieved the highest improvements, with a 7.5% increase in precision over PSO + DNN and 2.9% over Cuckoo + DNN. In terms of recall, Firefly + DNN outperformed PSO + DNN by 10.7% and Cuckoo + DNN by 6.3%. Furthermore, the F-measure of Firefly + DNN improved by 9.5% compared to PSO + DNN and 5.9% over Cuckoo + DNN. These refinements underscore the effectiveness of the Firefly Algorithm in selecting relevant features for defect prediction, resulting in more accurate, efficient models, and reliable. The study emphasizes the significance of feature selection in reducing overfitting, enhancing interpretability, and lowering computational costs. Overall, this research provides robust methods for improving software quality and reducing maintenance efforts through advanced defect prediction models, contributing significantly to the field of software engineering.