STRESS-AIDE: An AI-Powered Framework for Proactive Workload Stress Detection and Mitigation in Software Project Management

Prince Sharma

Computer Science and Engineering Nutan College of Engineering andResearch

Pune,India sharmaprince2659@gmail.com

Atharv Bhoite

Computer Science and Engineering Nutan College of Engineering andResearch

Pune, India atharvabhoite@gmail.com

Sandeep Hanamantgola Computer Science and Engineering Nutan College of Engineering and Research

Pune, India hanamantgolasandeep@gmail.com

Prof.Madhavi Patil

Computer Science and Engineering Nutan Maharashtra Institute of Engineering and Technology

Pune, India madhavi.patil@nmiet.edu.in

Arnav Kadam

Computer Science and Engineering Nutan College of Engineering andResearch

Pune, India arnavkadam777@gmail.com

Abstract—Software developers working in agile software development environments regularly face workloads that exceed their mental capacity limits. However, current project management (PM) systems do not provide any live tool for detecting and alleviating this stress. In this paper, we present STRESS-AIDE, an intelligent tool that utilizes the Trello REST API to extract live metadata about tasks, derives seven numerical attributes related to workload, and applies a machine learning-based classification model to categorize the personal stress of each user into one of three categories—Low, Medium, and High. Among the three different classifiers tested—Logistic Regression (the baseline), Random Forest, and XGBoost—XGBoost obtained the best performance with 89.0% accuracy and 87.5% macro F1- score. Proposed is a two-layer recommendation system that uses seven deterministic rules along with three score-driven recommendations from machine learning to provide ranked context- specific guidance on workload. Recommendations, predictions, and workloads are displayed in real-time using the React.js dashboard. User feedback is gathered through a human-in- the-loop framework, allowing the model to be incrementally retrained and customized according to individual stress metrics. The application is packaged using Docker Compose and achieves a latency of 42 ms while maintaining an average test coverage of 91% across 85 tests. STRESS-AIDE proves that the metadata from the project management tool is a sufficient signal in order to predict workload stress, thus linking task visibility to cognitive well-being awareness.

Index Terms—Workload stress prediction, XGBoost, Trello API integration, FastAPI, React.js, recommendation engine, human-in-the-loop machine learning, Docker, feature engineering, software project management.