Automatic detection of technical debt in large-scale java codebases: a multi-model deep learning methodology for enhanced software quality

Abstract Management of technical debt (TD) is crucial in long-term software projects for sustaining code quality. We proposed an effective deep learning-based approach to automatically detect and analyze self-admitted TD from large-scale Java codebases. Using a dataset consisting of over 55 million Java source files, we have designed several insightful machine learning models, including random forest, gradient boosting, long short-term memory, and gated recurrent unit, for making predictions about the presence and severity regarding TD. This proposed approach automates the risky component identification; therefore, one can manage TD proactively, thus reducing its costs and augmenting the overall project outcomes. Our results also confirm that these models have much increased detection accuracies of TD, thus giving a lot back to the software engineering domain.