Smart AI-Intensive Software Defect Predictions Based on Optimized Deep Feature Classification Using GWO-Code2Vec CNN Learning

Abstract

Software defect prediction plays a crucial role in improving software reliability, reducing maintenance costs, and enhancing software quality. However, traditional prediction methods often fail to accurately detect software defects due to non-relational dependencies among features, redundant datasets, and imbalanced data distributions. These limitations result in lower true positive rates and reduced predictive accuracy. To overcome these challenges, this research introduces an optimized Deep Learning–based Defect Prediction Framework that integrates advanced preprocessing, feature optimization, and classification mechanisms. Initially, Z-Score Logarithmic Transformation (ZSLT) is employed for preprocessing to normalize feature scales and eliminate noise in defect logs. To address data imbalance, the Clustered Synthetic Minority Oversampling Technique–Edited Nearest Neighbour (CSMOTE-ENN) method is applied, effectively balancing the dataset while preserving meaningful feature diversity. Subsequently, Grey Wolf Optimization–Deep Neural Network (GWO-DNN) is utilized for optimal feature selection, enabling the identification of highly correlated defect-related attributes. Finally, the Code2Vector–Graph Convolutional Neural Network (Code2Vector-GCNN) model is employed for deep feature classification, capturing both semantic and structural code representations to improve defect detection accuracy. Experimental results demonstrate that the proposed framework significantly outperforms conventional machine learning and deep learning models in terms of precision, recall, and F1-measure, providing an intelligent, adaptive, and high-accuracy solution for software defect prediction.