An AI-Based Approach for Accurate Fall Detection and Prediction Using Wearable Sensors

Falls are a paramount concern in elderly care and injury prevention, necessitating accurate and timely inter-ventions. This work contributes to transforming fall detection and prevention in elderly care by combining Convolutional Long Short-Term Memory (ConvLSTM) networks for real-time detection and Exponential Smoothing for early prediction. The proposed solution achieves an impressive Fl-score of 0.991 when tested on various public datasets of 81 subjects, showcasing its effectiveness. In addition to real-time detection, the approach introduces proactive prediction, forecasting falls before they occur and significantly reducing response time to 1100–1250 ms with an accuracy of 98.3%. This integration of real-time detection and early prediction addresses the gap in traditional systems, improving patient safety and lessening healthcare burdens.