DebtGuard: A Predictive Model For Managing Technical Debt In Agile Development

Technical debt (TD) in agile software development arises from rapid iterations prioritizing speed over quality, leading to increased maintenance costs and reduced sustainability. Effective TD management is crucial for agile teams to maintain software quality. This study proposes DebtGuard, a predictive framework to identify, quantify, and prioritize TD in agile projects, aiming to improve code quality and reduce maintenance effort while supporting iterative development. DebtGuard employs a Random Forest classifier, integrating static code metrics (e.g., cyclomatic complexity, code smells from SonarQube) and agile process metrics (e.g., sprint velocity, refactoring frequency). Data from 2,000 code modules across Apache Kafka, Spring Boot, and Elasticsearch were pre-processed via normalization, imputation, and recursive feature elimination. The model was trained with 10-fold cross-validation and evaluated against baselines like SonarQube and logistic regression over six months. DebtGuard achieved 89% accuracy, 86% precision, and 85% F1-score in predicting TD, outperforming baselines. It reduced maintenance effort by 24% on average, with a 26% reduction in Elasticsearch. Robustness was confirmed through visualizations, including ROC curves, precision-recall plots, and confusion matrices. DebtGuard offers a scalable, predictive TD management solution, enabling agile teams to balance delivery speed and quality. Its compatibility with tools like Jira enhances practical adoption. Limitations include reliance on labelled TD data and comprehensive logs. Future research could explore unsupervised learning, real-time CI/CD integration, and longitudinal studies to assess long-term impacts, advancing sustainable software engineering.