Fall Risk Assessment Using Pressure Insole Sensors and Convolutional Neural Networks

Falls prevention among the elderly community is regarded as one of the most critical public health topics in today’s aging society. Identifying the risk of falling in elderly individuals is considered the first step in prevention. In this study we present an alternative method of representing signal cyclostationarity as heat-map images and using convolutional neural network (CNN) with the ADAM optimization method to predict the risk of falling in 411 subjects over the age of 65. The study involved three different modes of walking: normal straight walking (MS), walking straight while calling out names of animals (MF), and walking straight while de-counting from the number 50 (MD). Data from the elderly participants were collected from wearable insole pressure sensors. Results obtained in this study showed improved prediction capability (increased accuracy by 6.8%) compared to traditional machine learning methods. In addition, the proposed method achieved improved results with reduced time in data collection as it requires the subject to perform one type of walking mode (MD) instead of three.