Fall Detection by 1D-CNN Based on Public Dataset Using a Wrist-Worn Smartwatch

Falls among the elderly population pose a significant threat, often leading to severe injuries, loss of independence, and even fatality. To address this critical issue, researchers have been exploring various approaches to develop effective fall detection systems. In this study, we present a novel fall detection algorithm that leverages the WEDA-FALL dataset, a comprehensive dataset that includes data from wrist-worn sensors capturing both simulated and real-world fall incidents involving elderly individuals. The key insight behind our approach is the observation that falls often exhibit distinct patterns in sensor data. To capture these patterns, we employ a one-dimensional convolutional neural network (1D-CNN) architecture, which has been shown to be well-suited for time-series data analysis. By training our model on the WEDA-FALL dataset, we were able to achieve an impressive accuracy of 94% in fall detection, surpassing the commonly reported accuracy levels of around 90% in the existing literature. The high performance of our algorithm can be attributed to the use of the WEDA-FALL dataset and the effectiveness of the 1D-CNN architecture in extracting and learning the characteristic patterns in the sensor data. Notably, our algorithm offers the key features of being quick, highly accurate, and cost-effective, making it a promising solution for practical deployment in real-world settings to improve the safety and well-being of the aging population.