Machine Learning-Based Gait Characterization Using Single IMU Sensor

Ambulation plays a vital role in the daily life activities of humans. However, people with neurological, gait, and locomotion disorders experience numerous obstacles in performing daily activities. To assist such a population, tracking and estimating gait parameters with minimum circuitry is one of the foremost steps. In the present research, a machine learning-based system has been developed to estimate the temporal gait characteristics, critical gait events, and gait phases. Moreover, the developed system required only a single inertial measurement unit (IMU) sensor. The performance of the developed system has been estimated for three classifiers, support vector machine (SVM), linear discriminant analysis (LDA), and K-nearest neighbor (KNN), over five healthy subjects. For gait events identification, the average classification accuracy depicted by SVM, LDA, and KNN classifiers are, $91.53 \pm 4.23 \%, 91.68 \pm 5.76 \%$, and $90.62 \pm 5.19 \%$ (p-value $\gt0.05$), respectively. Further, for gait phase estimation, the average performance is shown by SVM, LDA, and KNN classifiers, $94.40 \pm 5.14 \%, 97.03 \pm 3.20 \%$, and $91.94 \pm 4.52 \%$ (p-value $\lt 0.05$), respectively. The developed system provides a cost-effective solution for gait characterization.