A Hybrid Machine Learning Framework for Accurate Prediction of Software Maintainability Metrics

The growing complexity of modern software systems necessitates the accurate prediction of maintainability metrics to ensure software quality and reduce maintenance costs. While traditional machine learning techniques have shown promise, this study introduces a novel hybrid framework that combines feature engineering, classical machine learning algorithms, and ensemble methods to predict software maintainability metrics effectively. Our approach focuses on integrating widely-used software metrics, such as cyclomatic complexity, lines of code, and coupling, with external factors like developer activity and historical issue resolution rates. The proposed framework leverages models like Random Forests, Gradient Boosting Machines (GBM), and Support Vector Machines (SVM), with hyperparameter optimization using grid search techniques. A comprehensive evaluation of real-world software repositories demonstrates that our hybrid framework outperforms deep learning-based methods in terms of accuracy, interpretability, and computational efficiency. This work provides software engineers and project managers with a robust tool for assessing maintainability early in the development lifecycle, enabling more informed decisions for code quality improvements and project management.