Software Defect Prediction System Using Optimized Bi-Directional LSTM

Software defect prediction is critical for improving software quality and lowering maintenance costs by identifying potential errors early in the development cycle. In this article, we propose a Software Defect Prediction (SDP) system that uses an optimized Bi-Directional Long Short-Term Memory (Bi-LSTM) network to forecast flaws with high precision. The Bi-LSTM effectively captures both past and future contexts from sequential data, making it ideal for modeling complex interactions in software metrics. To enhance prediction performance, we implement optimization techniques to fine-tune the network's hyperparameters, such as learning rate, number of hidden layers, and batch size. These optimizations improve the learning process's efficiency and accuracy while mitigating overfitting. Our proposed method is tested on benchmark software defect datasets, achieving accuracy rates of up to 98%, precision scores reaching 95%, and F1-scores of 96%. These results demonstrate significant improvements over traditional machine learning models and standard LSTM networks, highlighting the effectiveness of combining Bi-LSTM with optimization strategies. This system provides a reliable solution for defect prediction, contributing to more robust software development practices.