Hybrid BERT, CNN-BILSTM Model for Detecting Self-Admitted Technical Debt in Software Development

Software developers frequently take shortcuts during the software development process, which they document as self-admitted technical debt (SATD) through source code comments. While SATD enhances transparency since developers acknowledge its presence through source code comments, it significantly increases software maintenance costs, elevates defect rates, degrades software quality and poses security vulnerabilities. To address these issues, we propose a state-of-the-art hybrid machine learning model that integrates BERT, CNN and BiLSTM architectures with attention for the automated detection of self-admitted technical debt. We use BERT for contextual embeddings, BiLSTM for capturing sequential dependencies and CNN to capture local patterns. Employing an experimental approach, our model achieved an average F1-score of 0.78 with attention, an improvement when compared with our baseline models. The contribution of this paper is a high-performing hybrid machine learning model for the automatic detection of technical debt leading to better software quality and low maintenance cost.