Fall Risk Prediction in Older Adults Through Daily Activity Monitoring- A DBN-Based Approach

Falls in older adults continue to be a significant health concern despite advancements in healthcare and emerging technologies. Recently, fall-related research in older adults has attracted considerable attention from researchers worldwide, with a focus on developing high-performance methods for monitoring, predicting, and preventing falls. Enhanced performance can be achieved by understanding the features pertinent to the problem. This study aims to develop an intelligent fall prediction model that foresees future falls in older adults by monitoring their Activities of Daily Living (ADL) and identifying abnormalities through continuous monitoring. The proposed fall risk prediction technique utilizes a Deep Belief Network (DBN) based model, incorporating the Contrastive Divergence technique for pre-training, neural network-based fine-tuning, and Adams Optimizer to minimize the loss function. The evaluation of the proposed model shows an accuracy of 93.33%, a specificity of 100%, and a sensitivity of 92.86% when compared to the Morse Falls Scale (MFS). The results indicate that advanced deep learning techniques benefit older adults, facilitating the early prediction of fall risk and its associated severity, thereby reducing the risk of falls.