XAI driven software defect prediction using adaptive fature engineering coupled with autoencoder and multi-layer perceptron: an empirical study

Software defect prediction is essential for ensuring the reliability and robustness of software prototypes during development. The current study has proposed a novel strategy that integrates explainable artificial intelligence (XAI) techniques with adaptive feature engineering and autoencoder neural networks to improve defect prediction accuracy and interpretability. Adaptive feature engineering processes input data to identify critical features, while autoencoders handle non-linear datasets, reduce noise, and generate meaningful latent representations by learning underlying data patterns. A Multi-Layer Perceptron (MLP) is employed to classify code snippets and localize defects, leveraging its ability to manage complex data patterns and diverse input features. To enhance transparency and trust in model predictions, the XAI component provides insights into feature significance and classification outcomes. Empirical evaluations were conducted on the PROMISE dataset, with performance assessed using metrics such as sensitivity, specificity, F1-score, and the Matthews correlation coefficient. The proposed approach has demonstrated superior accuracy compared to other defect prediction methods, highlighting its effectiveness in identifying defective code snippets and enhancing software quality.