A comparative study of deep learning techniques in software fault prediction

Abstract

Software fault prediction (SFP) is an important approach in software engineering that ensures software quality and reliability. Prediction of software faults helps developers identify faulty components in software systems. Several studies focus on software metrics which are input into machine learning models to predict faulty components. However, such studies may not capture the semantic and structural information of software that is necessary for building fault prediction models with better performance. Therefore, this paper discusses the effectiveness of deep learning models including Deep Belief Networks (DBN), Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), and Long-Short Term Memory (LSTM) that are utilized to construct fault prediction models based on the contextual information. The experiment, which has been conducted on seven Apache datasets, with Precision, Recall, and F1-score are performance metrics. The comparison results show that LSTM and RNN are potential techniques for building highly accurate fault prediction models.