Binary and Ternary Human Gait Phase Classification Using Machine Learning Algorithms

Abstract

Accurately identifying gait phases offers vital in-sights into medical diagnostics, requiring precise biomechanical assessment. Despite various machine learning (ML) methods, achieving high precision is challenging. This study uses ML techniques to classify the gait phases of 100 individuals (average age 41.91 ± 5.3 years). Classification algorithms considered in this work are: k-Nearest Neighbor (k-NN), Logistic Regression (LR), Decision Tree (DT), Random Forest (RF), Support Vector Machine (SVM), and Naive Bayesian (NB). The performance of these algorithms is evaluated using two training methods while classifying binary (stance/swing) and ternary (stance-I1stance-Illswing) subphases. In the first method, 80% of the data from binary or ternary phases is randomly selected for training and the rest for testing. In the second method, data is divided by participants, with 80% used for training and 20% for testing. The RF algorithm consistently outperforms others in both training methods, achieving 99.5% (first method) and 99.9% (second method) accuracy in binary classification and 99.4% (first method) and 99.5% (second method) in ternary classification. Despite the reduced performance while transitioning from bi-nary to ternary classification, consistency is maintained for all ML algorithms. Moreover, the second training method is more effective for classifying binary and ternary gait phases.