Enhancing Software Co-Change Prediction: Leveraging Hybrid Approaches for Improved Accuracy

Abstract

Accurate prediction of co-changes in software systems is crucial for efficient development and maintenance, especially as systems grow in complexity. While deep learning-based approaches have shown promise, they often struggle with diverse and complex data. In this paper, we present a novel hybrid approach that combines traditional software engineering methods with deep learning techniques to improve co-change prediction accuracy. Our approach leverages software metrics and deep learning models, incorporating the unique characteristics induced by naming conventions, such as PascalCase and camelCase, used by developers for naming consistency. By utilizing char n-gram embedding and sub-token context, we enrich the vector representations of source file names, capturing relationships and dependencies between files. We comprehensively evaluate our hybrid approach using three open-source software projects. The findings of this study have significant implications for the development of more effective software co-change prediction tools and techniques, enabling better decision-making in software development and maintenance processes. Our approach outperforms traditional software engineering methods and deep learning-based approaches, demonstrating its potential to significantly improve software development and maintenance efficiency.