An Efficient Dual Attention Graph-based Recurrent Optimization Model for Software Defect Prediction

Abstract

Software defect prediction plays a critical role in improving software quality by identifying faulty modules early in the development process. This study aims to address the limitations of existing machine learning and deep learning models, such as their inability to effectively capture structural dependencies, temporal evolution, and high-dimensional feature interactions. To achieve this, a novel model named Dual Attention Graph-based Recurrent Optimization (DAGr-RO) is proposed. The model integrates graph-based learning to represent inter-module relationships, dual attention mechanisms to emphasize important features and structural connections, and recurrent neural networks (LSTM/GRU) to capture temporal patterns across software versions. Additionally, an optimization strategy is employed to dynamically update model parameters and enhance prediction performance. The proposed approach is evaluated using a Kaggle software defect dataset and compared with baseline models including RNN, CNN, Random Forest, and XGBoost. Experimental results demonstrate that DAGr-RO achieves superior performance with an accuracy of 94.5%, precision of 93.2%, recall of 92.6%, and F1-score of 92.9%, outperforming all comparative models. These findings confirm that the integration of graph learning, attention mechanisms, and temporal modeling significantly improves defect prediction. The study concludes that DAGr-RO provides an effective, scalable, and reliable solution for real-world software defect prediction tasks.