Walking-Age Estimator Based on Gait Parameters Using Kernel Regression

Human gait motions differ depending on the age of the person. Previous studies have estimated age categories of walkers or have used age analysis for security or commercial surveillance purposes using images. However, few studies have estimated age from gait parameters alone. We estimated the age of people using kernel regression analysis based on their height, weight, and representative gait parameters, i.e., walking features that are interpretable with relative ease. Samples were obtained from 75 Japanese women aged 20–70 in a database. Through a variable selection based on sensitivity analysis, the established model estimated the ages of the women with a correlation coefficient of 0.78 with their actual ages, and the mean absolute error was 9.99 years. The sensitive variables included the minimum foot clearance, body weight, walking velocity, step width, and stride length. Estimation errors were significantly greater for elderly adults than for young people. Specifically, the mean absolute error for people in their 20s was 7.4 years, whereas that for those over 60 was 13.1 years. The proposed method uses gait parameters that can be measured with wearable devices, such as inertial measurement units; therefore, it offers an accessible approach to estimating a walker’s age with moderate certainty and promoting healthcare awareness in daily life.