<![CDATA[Flexible workers operating under remote, hybrid, and freelance schemes face burnout risks that are difficult to detect early due to irregular work patterns and blurred work-time boundaries. Conventional burnout monitoring relying on manual surveys is static and lacks sensitivity to the dynamics of workers' psychological changes. This study aims to develop a machine learning-based burnout prediction system for flexible workers capable of providing real-time risk predictions accompanied by personalized prevention recommendations. The method employed is Random Forest Classifier using a dataset from Kaggle titled "Mental Health & Burnout in the Workplace" encompassing 5.000 observations. System development follows the Agile approach and is implemented through a Streamlit-based web application. Preprocessing stages include binary label transformation, data leakage elimination, one-hot encoding, class imbalance handling using SMOTE, and stratified split with a 90:10 ratio. The Random Forest model is configured with 800 trees, max_depth of 20, and other optimal hyperparameters. Evaluation results demonstrate that the model achieves 87% accuracy with precision of 0.89, recall of 0.91, and F1-score of 0.90 for the burnout class. Feature importance analysis identifies CareerGrowthScore, StressLevel, and ProductivityScore as dominant factors. The system provides real-time predictions with latency <2 seconds and prevention recommendations tailored to individual risk profiles. This research contributes a practical solution for self-monitoring mental health among flexible workers and provides organizations with an instrument for monitoring remote workforce well-being. Black-box testing validates that all functionalities operate according to specifications.]]>
