IoT-Based Lifting Posture Detection using Random Forest for WMSD Prevention

Work-Related Musculoskeletal Disorders (WMSDs) from improper lifting techniques remain a major occupational health concern globally. This study develops an Internet of Things (IoT)-based lifting posture detection system using Random Forest classification to enable real-time monitoring and early WMSD prevention. The system integrates two MPU6050 inertial sensors placed on the thoracic and lumbar regions, a BMP280 pressure sensor attached to the load (15-liter water gallon), an ESP8266 microcontroller, and a Flask-based web application with buzzer feedback. Data were initially collected from 24 participants, with 20 included in the final analysis after excluding incomplete datasets, generating 53 engineered features from acceleration, gyroscope, pressure, and orientation signals. The Random Forest model was evaluated using leave-one-subject-out cross-validation and achieved an overall accuracy of 83.32%, with the correct posture class showing higher recall (0.87) and F1-score (0.85) compared to the incorrect class (recall 0.79, F1-score 0.81). Feature importance analysis identified thoracic and lumbar pitch angles as key predictors of safe lifting biomechanics. Real-time system testing across 40 activities yielded a 97.5% prediction accuracy with an average confidence of 86.7%, demonstrating the system's practical viability. While the model shows promise for automated posture monitoring, performance variability across subjects (68.62%–96.20%) indicates the need for larger, more diverse training datasets and further algorithm refinement. This research establishes a foundation for IoT-based ergonomic intervention systems that provide immediate auditory feedback to reduce WMSD risk in occupational settings.