"Deep Learning for Fall Risk and Health Detection in Individuals with Multiple Sclerosis: A GoogLeNet-Based Approach with UWB Dataset"

Abstract

- Falls are a prevalent cause of injuries among seniors, particularly within indoor environments such as homes, nursing homes, senior living communities, and care facilities. Recognizing the paramount importance of predicting and understanding the actions of older individuals during falls, this study explores the application of deep learning, specifically the GoogleNet model, in predicting fall risks. The investigation encompasses both seniors and individuals with Multiple Sclerosis (PwMS), utilizing a dataset derived from Ultra-Wideband (UWB) technology.The research addresses the significant risk that falls pose to the well-being of both persons with Multiple Sclerosis and the general aging population. Leveraging UWB technology and the deep learning capabilities of the GoogleNet model, the study seeks to develop an accurate and reliable predictive system for fall risk assessment.The methodology involves preprocessing UWB data, organizing it into labeled folders, and applying the continuous wavelet transform (CWT) to generate a time-frequency representation of the data. The GoogleNet model, pretrained on a diverse dataset, is then adapted for transfer learning to suit the specific task of fall risk prediction. Modifications include introducing a new classification output layer and adjusting the fully connected layer to accommodate the required output classes.Training the model utilizes the UWB dataset, with probabilistic gradient descent and mini-batch updates. Validation on a separate dataset monitors training progress, and the final model is tested on validation data. The predictive system is assessed for both persons with Multiple Sclerosis and healthy individuals, acknowledging the unique challenges faced by each group.