A Systematic Synthesis of Software Maintainability Paradigms: From Static Metrics to Predictive Intelligence

Abstract

The high cost of maintenance and the demand for sustainable systems drive the need for a deeper understanding of software maintainability. Therefore, this study aims to map the latest trends, from evaluation metrics to solutions for code smells and technical debt. We applied the Systematic Literature Review (SLR) method based on the PRISMA guidelines to critically examine eight primary studies. The extracted data were then analyzed thematically to find common threads across the studies and identify underexplored areas. Our analysis highlights an interesting duality: classic metrics remain a relevant foundation, but the dominance of machine learning in predicting maintainability is increasingly undeniable, despite the accompanying imbalanced data constraints. The literature consistently confirms that code smells and technical debt are major bottlenecks to code quality, which can now be mitigated through automated refactoring strategies and prioritizing the handling of bad smells. Another crucial finding is the vital role of historical data on code evolution for prediction accuracy, as well as the emergence of a new perspective linking maintainability to energy efficiency. This study enriches the literature by consolidating empirical evidence, which not only validates previous theories but also identifies remaining research gaps. From a practical perspective, the results of this analysis can serve as a strategic reference for practitioners and academics in sharpening maintainability predictions and mitigations—key steps to reduce costs and improve software quality. As a future agenda, we recommend focusing research on developing more adaptive predictive models and deeper investigations into non-technical factors. 

Keywords: code smells, green IT, machine learning, technical debt, software maintainability.