Deep Learning-Based Software Fault Prediction Using Product and Process Metrics

Abstract

Software Fault Prediction aims to identify software faults in advance, enhancing performance. Machine learning techniques are used to predict software defects. While early approaches used product metrics, later advancements integrated process metrics to enhance predictive capabilities. Deep learning has emerged as a powerful technique for fault prediction. However, current research has primarily focused on utilizing product metrics. The effectiveness of deep learning models when applied to combined or process metrics has not yet been investigated. The novel contribution of this research is the integration of process metrics and product metrics, leveraging deep learning models to improve fault prediction. The study included two experiments: (i) using deep learning models such as Convolutional Neural Network, Long Short-Term Memory (LSTM) and Bidirectional Long Short-Term Memory (BiLSTM) network to evaluate product-only metrics with combined metrics (product + process), and (ii) comparing the performance of deep learning models with conventional machine learning models (k-Nearest Neighbors (k-NN), Naive Bayes, and Logistic Regression) based on combined metrics. Experimental results show that fault prediction performance is enhanced by combining process metrics with product metrics. Deep learning models achieved an accuracy of up to 0.93 across five benchmark datasets, with precision, recall, and F1-scores of 0.93, 0.98, and 0.95, respectively. The combined metrics enhanced performance across all parameters when compared to product-only metrics. Additionally, deep learning models such as Convolutional Neural Networks and Bi-Directional Long Short-Term Memory performed better than machine learning techniques, demonstrating greater stability and efficiency in identifying complex patterns. Thus, this research demonstrates the effectiveness of integrating metrics within deep learning approaches for fault prediction.