A comparison of some soft computing methods for software fault prediction

The main expectation from reliable software is the minimization of the number of failures that occur when the program runs. Determining whether software modules are prone to fault is important because doing so assists in identifying modules that require refactoring or detailed testing. Software fault prediction is a discipline that predicts the fault proneness of future modules by using essential prediction metrics and historical fault data. This study presents the first application of the Adaptive Neuro Fuzzy Inference System (ANFIS) for the software fault prediction problem. Moreover, Artificial Neural Network (ANN) and Support Vector Machine (SVM) methods, which were experienced previously, are built to discuss the performance of ANFIS. Data used in this study are collected from the PROMISE Software Engineering Repository, and McCabe metrics are selected because they comprehensively address the programming effort. ROC-AUC is used as a performance measure. The results achieved were 0.7795, 0.8685, and 0.8573 for the SVM, ANN and ANFIS methods, respectively.