Pre-Training of an Artificial Neural Network for Software Fault Prediction

Software fault prediction is one of the significant stages in the software testing process. At this stage, the probability of fault occurrence is predicted based on the documented information of the software systems that are already tested. Using this prior knowledge, developers and testing teams can better manage the testing process. There are many efforts in the field of machine learning to solve this classification problem. We propose to use a pre-training technique for a shallow, i.e. with fewer hidden layers, Artificial Neural Network (ANN). While this method is usually employed to prevent over-fitting in deep ANNs, our results indicate that even in a shallow network, it improves the accuracy by escaping from local minima. We compare the proposed method with four SVM-based classifiers and a regular ANN without pre-training on seven datasets from NASA codes in the PROMISE repository. Results confirm that the pre-training improves accuracy by achieving the best overall ranking of 1.43. Among seven datasets, our method has higher accuracy in four of them, while ANN and support vector machine are the best for two and one datasets, respectively.