Software Fault Prediction Using an Optimised Feature Selection Process Based on a Genetic Algorithm

Software Fault Prediction (SFP) is crucial for preemptively identifying software faults. This research tackles challenges in SFP, addressing class imbalance, significant metrics, and feature selection. To counter class imbalance, the Random Over-sampling method is employed, enhancing model predictive capacity for both faulty and non-faulty instances. Software metric categories like size, cohesion, complexity, coupling, and documentation are examined to identify influential metrics. Feature selection is optimized using a Modified Genetic Algorithm (GA), reducing dimensionality while maintaining efficiency. Experiments on a diverse open-source dataset demonstrate that this approach substantially improves prediction accuracy compared to traditional methods. This study introduces a comprehensive framework for robust SFP models. By handling class imbalance, recognizing significant metrics, and implementing effective feature selection, the proposed approach empowers practitioners to build more accurate fault prediction models, enhancing software quality and reliability.