Abstract
Most software development projects use Agile project management which emphasizes rapid iteration and adaptability. Agile IT projects work within fast-changing and volatile environments where risks can change at any time, managing risks in such setups is often difficulty. Project failures have been linked to poor risk management, underscoring the need to explore more effective methods to manage risks. Current AI-enabled risk management tools use static models with limited capabilities of being truly adaptive to evolving risks. The study investigates which machine learning model can effectively support real-time risk monitoring in Agile projects and compares the performance of different model types under identical conditions.
Through using a quantitative experimental design, the study compared different ML models under identical conditions using four publicly available software project datasets. Models were grouped into three categories: traditional classifiers, ensemble methods and adaptive streaming algorithms. Risk monitoring was simulated through Jira APIs and performance was evaluated using metrics such as accuracy, precision, recall, F1-score, ROC-AUC, and confusion matrix analysis.
The results show that ensemble models are best for real-time risk monitoring in Agile projects, within a standardised, untuned setup. Algorithms such as Random Forest and Bagging achieved the highest F1-scores and lowest false negative rates. Traditional models such as Decision Trees also performed well and provided interpretable baselines. Adaptive streaming models showed weaker overall performance though they showed potential in datasets where data distributions changed more strongly over time.
The study proposes a conceptual framework for adaptive, real-time risk monitoring in Agile IT projects that integrates machine learning models with Agile project tools and dashboards. Through this framework the outputs results prove ensemble models to be suitable AI tools for continuous monitoring in Agile workflows. When embedded into dashboards, ensemble models provide a backbone for automated risk alerts in Agile sprints.