Experimenting Software Defect Density Prediction using Data Attention Point and Deep Learning

الملخص

This research focuses on addressing the issue of predicting software defect density by leveraging advancements in machine learning and deep learning techniques. By integrating attention mechanisms, like self attention and cross attention into a deep learning framework we introduce a model named AttDeep. This model aims to enhance the accuracy of software defect prediction. The attention mechanisms help in reflecting the content across all data parts thereby improving its capability to identify patterns and anomalies that could signal defects. Through experiments using datasets from several repositories our findings indicate that AttDeep not only delivers performance with reduced loss and mean absolute error (MAE) but also exhibits enhanced generalization compared to models without attention mechanisms. These results highlight the effectiveness of incorporating attention mechanisms into deep learning for software defect prediction suggesting avenues, for research and practical applications by integrating attention mechanisms into machine learning algorithms.