Simulation Annealing Based Bacterial Foraging Optimization with Support Vector Machine for Predicting Potential Defects in Software

In recent times, software prediction systems have gained popularity as they enhance software reliability by effectively recognizing potential bugs in a code. However, the predictions made by the existing models are vulnerable to redundant or irrelevant features, which reduce the overall performance. Therefore, this research proposes a Simulation Annealing based Bacterial Foraging Optimization (SABFO) with Support Vector Machine (SVM) for predicting potential defects in software. The SABFO helps determine relevant features based on the exploring capability of SA. The SA enables the algorithm to avoid local optima by including a random component to neighborhood search, while the BFO provides an effective mechanism for local search based on chemical attractiveness and reproduction rules that allow the algorithm to achieve the best feature subsets. The PROMISE dataset is preprocessed by Z-score normalization, which is used to scale the input features between the range of 0 and 1. The measures of F1-score, G-measure, false alarm (pf), and Matthews Correlation Coefficient (MCC) are used to assess the SABFO-SVM model's performance. The SABFO-SVM achieves a satisfactory performance with 0.8247, 0.8549, 0.2174, and 0.5619 of F1-score, G-measure, pf and MCC, respectively.