A hybrid chaos-based feature selection and machine learning for improved software defect prediction

One of the main challenges in software engineering is software defect prediction, which aims to improve software quality by proactively identifying probable faults. By identifying defect-prone modules before testing, this procedure reduces testing costs and time. Machine learning is an excellent approach to classifying software modules as potentially defect-prone or defect-free. Nonetheless, certain features in the dataset may affect the precision and accuracy of the defect predictors. Ineffective and redundant features in a training dataset will reduce the performance of the resulting defect predictor. In this study, a hybrid Chaos-Based Feature Selection and Machine Learning Framework (CHAFEM) has been proposed to enhance software defect prediction. By concentrating on the selected crucial features, software module classifiers can be more accurate and precise. A chaos-based Shuffled Frog Leaping Optimizer has been developed as a feature selector. The extracted optimal features were used to train different machine learning algorithms and create an effective defect predictor. The effectiveness of the proposed defect predictor was evaluated on five real-world and standard datasets. The results show that, among the 21 features in the raw dataset, intrinsic complexity, lines of code, the total number of operators and operands, and the number of comments in the source code are the most influential. This paper highlights that integrating the binary Shuffled Frog Leaping Optimizer with machine learning algorithms significantly improves accuracy, F1 score, precision, and sensitivity in defect prediction.