Software Defect Prediction Based On Deep Learning Algorithms : A Systematic Literature Review

Software bug prediction (SDP) techniques identify bugs in the early stages of the software development life cycle through a series of steps to produce reliable and high-quality software. Deep learning techniques are widely used in SDP, which can produce accurate and exceptional results in different fields.The study aims to systematically review models, techniques, datasets, and performance evaluation metrics to gain a complete understanding of current methodologies related to SDP, and the use of DL in software defect prediction research between 2019 and 2024. A comprehensive review of studies in this area was completed to answer the research questions and summarize the results from the initial investigations. 30 primary studies that passed the systematic review quality assessment of the studies were used. However, the six most common evaluation metrics used in SDP were confusion matrix, Scoar-1F, recall, precision, accuracy, and area under the curve (AUC). The top three DL algorithms used in building SDP models and used in predicting software bugs were convolutional neural network (CNN), long-short-term memory (LSTM), and bidirectional LSTM. We conclude that the application of deep learning in SDP remains a challenge, but it has the potential to achieve better prediction performance. Future research directions focus on improving these models and exploring their applications across diverse programming environments