Predicting Software Reusability Using Classification Techniques

This study predicts software reusability at the class level using machine learning and a dataset of 65 Java applications. The reuse rates were calculated using cohesion, coupling, complexity, inheritance, documentation, and size metrics. The data was segmented by project size (all, small, medium, and big). Several classifiers were tested once the labels were changed to multi-class labels (low: 0, moderate: 1, high: 2). On large projects, the voting model was 93% accurate. The results demonstrate how well machine learning predicts software reusability and provide recommendations for increasing software quality measures.