The Effect of Sensor Placement for Accurate Fall Detection based on Deep Learning Model

Abstract

The development of inertial sensor technology and the growing utilization of wearable electronics (such as smartwatches, smart bands, and other intelligent gadgets) have facilitated the advancement of studies into automated Fall Detection Systems (FDSs). In the last decade, there has been significant scientific interest in maintaining FDSs. Focused on assessing the data acquired by wearable inertial sensors, machine learning (ML) techniques have demonstrated high efficacy in distinguishing falls from typical motions or activities of daily living (ADLs). In most research, unfortunately, the effectiveness of machine learning classifiers was constrained by feature extraction and selection processes that relied on human-made decisions. Recently, deep learning (DL) model findings s how their effectiveness for FDS. One of these effective DL models is the ResNeXt model, a deep neural network that operates based on convolutional layers with aggregated residual transformation. This study investigates the influence of sensor placement on various body locations for the fall detection issue. The ResNeXt model was assessed and compared to other baseline deep learning algorithms using the public UMAFall dataset for fall detection. Employing sensor data on waist location, the suggested model attained the most significant classification ac curacy of 97.275% when classifying falls.