Software Fault Prediction Using Combinations of Code Smells, Code Metrics, and Code Smell Metrics With Ensemble and Deep Learning

Abstract Code smells are the structural characteristics of the software under development that indicate poor code choices and can cause errors or failures in the software and they can degrade the software maintenance and evolution processes. Software fault prediction (SFP) helps predict the probability of the existence of software faults utilizing code characteristics and metrics. Code smells-based datasets are based on a variety of code measures such as information about the presence of certain code smells or a combination of code smells with code metrics and code smell metrics with code metrics. We investigate the effectiveness and usefulness of these different combinations for performance evaluation and improvement of SFP models. We label the unlabeled datasets using clustering and pseudo-labeling techniques. We implement models considering ensemble methods and deep learning algorithms and compare performance. We use k-fold cross-validation and our results outperform existing benchmark studies. We conclude that code smells-based software defect prediction has optimal accuracy and precision.