AI-Driven Frameworks for Efficient Software Bug Prediction and Automated Quality Assurance

Abstract

This paper introduces an AI-driven framework designed to enhance software bug prediction and automated quality assurance. The framework leverages advanced machine learning and deep learning techniques, particularly Deep Belief Networks, to effectively predict bug-prone areas within software systems. The methodology begins with the collection of comprehensive datasets, which include essential features such as code complexity, commit logs, bug reports, and developer activity. These data are pre-processed to handle missing values, remove duplicates, and normalize for consistency, ensuring the dataset is ready for model training. Feature extraction using Wavelet Transform is then applied to capture both fine-grained and broad patterns within the data. The core of the model consists of Restricted Boltzmann Machines in the hidden layers of the DBN, which learn hierarchical representations from the input data. The output layer classifies the software as either buggy or non-buggy based on the learned features. The model’s performance was evaluated and showed promising results in predicting software defects, offering valuable insights into areas requiring attention. The study compares the AI-driven approach with traditional defect prediction methods, demonstrating higher adaptability and accuracy in dynamic environments. The results suggest that further refinement, real-time data integration, and hybrid model development could further enhance the model’s predictive capabilities for software validation and quality assurance processes.