Daily Locomotor Movement Recognition with a Smart Insole and a Pre-defined Route Map: Towards Early Motor Dysfunction Detection*

Abstract

Motor dysfunction, a well-known early sign of neurodegenerative diseases, is occurring to seniors at a growing rate and affects their physical capability of independent living if not treated effectively. The symptoms of motor dysfunction are hard to notice at early stages and can deteriorate in the long term. Thus, it is desirable to detect motor function changes in daily life in a noninvasive manner. This paper aims to accomplish this goal by proposing a method to auto-recognize nine types of daily activities from continuous movements with the use of a smart insole and a pre-designed route map. The route map creates a semi-controlled environment to help the subjects take actions comfortably and behave in experiments as they do in real life. The nine types of highly similar activities are selected from the motor examination and the balance evaluation system test. Preliminary experiments were done with four subjects with controlled and uncontrolled data collection. Four supervised machine learning classifiers are evaluated and compared for classification performance with a 2s window and different overlaps. With regards to the performance and robustness of classifiers, the Random Forest classifier trained with Mix Dataset shows the best results with an averaged classification accuracy of 98.19% in model training, 92.67% in cross-validation and 83.87% in prediction. The results show that it is feasible to recognize these nine activities from daily locomotor movement and further extract parameters of interest from activity periods for early motor dysfunction detection.