Sudden Fall Detection and Prediction Using AI Techniques

Fall prediction is a critical process in ensuring the safety and well-being of individuals, particularly the elderly population. This paper focuses on the development of a fall detection and prediction system using wearable sensors and machine learning algorithms. The system issues an alarm upon predicting the occurrence of falling and sends alerts to a monitoring centre for timely assistance. Wearable sensor devices, including Inertial Measurement Units (IMUs) equipped with accelerometers, gyroscopes, and magnetometers are utilized for data collection. UPFALL, a comprehensive online freely available dataset, had been utilized for training and verifying the proposed system. Several machine learning algorithms, such as K-Nearest Neighbours (KNN), Random Forest, Support Vector Machine (SVM), and Gradient Boosting, are implemented and evaluated. Among these algorithms, KNN demonstrates the highest effectiveness for fall detection having an accuracy of 93.5%. Furthermore, deep learning models, including Gated Recurrent Units (GRU), Long Short-Term Memory (LSTM), and Convolutional Neural Network (CNN), are employed. The GRU model exhibits superior performance among the deep learning approaches by having the least train and test loss of 0.219 and 0.267 respectively. An early fall prediction function is incorporated by establishing a threshold selection process based on logical analysis. The maximum voting concept is employed to determine the optimal threshold.