Document Type


Source Publication Title

PETRA 2022


This paper presents a Machine Learning approach using sensor data from a Smart Floor aimed at addressing a substantial health problem among the elderly population, namely falls. Studies show that one-third of community-dwelling people over age 65 will experience one or more falls each year. Balance and walking patterns are useful indicators to determine the fall risk and are influenced by several parameters and conditions. The Tinetti test is widely used to assess the gait and balance in elder adults to determine the perception of balance and stability during daily activities and fear of falling. It is considered a good indicator of the fall risk. In this research, we aimed to provide a new way for fall risk reduction and early detection of the onset of chronic health conditions by creating a Machine Learning model for predicting Tinetti scores based on foot pressure data arising in common everyday activities. The goal is for this to help improve eldercare through constant monitoring. This paper focuses on designing algorithms to extract balance parameters from standing instances arising during normal everyday life and to build individualized models capable of differentiating normal or abnormal patterns for an individual. A variety of time and frequency domain parameters are built based on the center of pressure (CoP) values obtained from time-series data from a pressure monitoring floor sensor. A classification model is build using a support vector machine for distinguishing 30 individuals based solely on these balance parameters. Further, using these parameters, a regression model is built to predict the balance and the Tinetti score of an individual which is used to predict the fall risk. This novel approach for Tinetti score prediction by just using balance analysis could be used in isolation or in combination with other assessments to assess health changes and to prevent falls before they happen.

Publication Date





Creative Commons Attribution 4.0 International License
This work is licensed under a Creative Commons Attribution 4.0 International License.