Prediction and identification of multiple sclerosis patients from gait features using machine learning algorithms and ensemble methods in embedded sensors environment

Multiple sclerosis is a chronic autoimmune disease prevalent among mid-aged to younger adults. The objectives of implementing advanced computational techniques in gait feature analysis are to refine the analytical process, improve the accuracy of gait assessment and understanding human locomotion patterns to predict and identify the multiple sclerosis patient using logistic regression, Support Vector Machine and K-Nearest Neighbours classifiers. We aim to automate the analysis of gait feature data collected from various sensors to expedite the prediction and identification of multiple sclerosis. This automation facilitates the early detection of subtle deviations in gait dynamics, which could signify neurological impairments. Logistic regression, Support Vector Machine and KNN models are capable of processing intricate gait data to discern meaningful patterns and characteristics indicative of multiple sclerosis conditions with minimum number of gait features. This paper introduces new gait features to improve the model’s performance and differentiate the multiple sclerosis from healthy control group. KNN demonstrates superior accuracy (96.55%) compared to Support Vector Machine (90.05%) and logistic regression (86.35%). KNN with bagging achieved the highest prediction accuracy (94.4%), while logistic regression with bagging had the lowest (83.3%). This paper proposes an ensemble learning method that uses weight averaging to get combine the predictions of above models with ensemble accuracy of 90.12%. A new index called match index is created for the identification of multiple sclerosis.