Clustering Mechanical Power of Lower Limb Joints During Gait Using K‑Means Clustering

ABSTRACT Given the inherent essence of gait, there is potential to analyse this data using artificial intelligence. Twenty‐eight young individuals completed a walking task. Data points were categorised using a machine learning technique, k‐means clustering. The movements of the hip, knee and ankle joints were compared across different clusters. Determined the number of clusters ( k ) using the elbow criterion and mean silhouette analysis across k = 2–6, which consistently favoured k = 3. Then compared peak joint power (production/absorption) between clusters using one‐way ANOVA, reporting statistically significant differences (p < 0.05). Angular velocity and joint moments significantly influence the distribution of subjects in clusters. Cluster 3, the smallest, exhibits the highest mechanical power. The mechanical power of the ankle and hip joints affects clustering patterns, while the second group contains few participants with heterogeneous power distribution across joints (e.g., at steady speed, Cluster 2 showed high ankle production and knee absorption without a single joint dominating overall) among joints. The knee joint is the primary determinant in the clustering process and the hip joint plays a minor role in power generation and absorption at preferred speeds. Significant differences in production and absorption phases of gait in preferred and steady speed conditions, except for the hip joint.