A Lightweight Bug Prediction Method for Medium-Scale Software Projects

Abstract

Software defect prediction remains difficult for medium-scale development teams with limited historical data. Complex learning models often require large datasets and extensive tuning. This paper presents a lightweight bug prediction method based on code metrics and change frequency analysis. The method was evaluated on six open-source projects containing between 40,000 and 120,000 lines of code. Compared with traditional rule-based approaches, prediction accuracy improved by 9%–14% under cross-version validation. The proposed approach requires minimal parameter adjustment and can be deployed in resource-constrained development environments.