This research is direct at developing an e-health approach for preventing the Musculoskeletal Disorders in construction through monitoring injury risk. The proposed approach leverages activities recognition through motion capturing techniques to monitor, assess, and reduce injury risk. More specifically, the authors used the Inertia Measurement Units, a motion-sensing tool to develop a concept for a wearable motion-data capturing prototype. The captured motion data was analyzed using data-driven, Machine Learning techniques to identify injury-prone activities. Instead of adopting a generic recognition model, a novel rapid model training process was investigated to configure a user-specific activity recognition model, aiming at improving the recognition accuracy with reduced computational effort. The customized model was based on the optimal configuration of data segmentation window size, feature sets, and classification algorithms for a specific user's activity data. The feasibility study of the proposed approach has shown the personalized model achieved an average overall recognition accuracy of 0.81 and 0.74 for two sets of activities. The recognition model's operation time was also reduced to under 0.01 seconds. The proposed approach can help to address the scalability challenge for data-driven activity recognition, and further, improve the effectiveness and practicality for proactive injury prevention on construction job site.