Gait classification in individuals with unilateral transfemoral amputation using random forest and k-means clustering

Cluster analysis has been recently applied to categorize gait patterns in individuals with unilateral transfemoral amputation (uTFA). However, conventional clustering methods largely rely on experiential knowledge of gait analysis, lacking a scientific foundation for feature selection. The aim of this study was to investigate if gait patterns could be classified using random forest and k-means clustering in individuals with uTFA. Spatiotemporal data and vertical ground reaction force (vGRF) were collected using an instrumented treadmill from twelve individuals with uTFA and twelve age-matched non-disabled individuals participated. Absolute symmetry index (ASI) was obtained and normalized. These parameters served as inputs for a random forest model to assess their importance. K-means clustering was applied to determine the optimal number of clusters according to Silhouette Score and the Elbow Method. Differences in demographic, spatiotemporal, and ASI parameters among clusters were assessed using One-way ANOVA and independent-sample Kruskal-Wallis tests. Random forest model revealed that swing phase and single limb support duration time symmetries were most significant for distinguishing individuals with uTFA from non-disabled. The k-means identified three distinct clusters: cluster 1 exhibited the lowest symmetry with the shortest prosthetic single limb support duration; cluster 2 displayed the highest symmetry with the longest prosthetic single limb support duration and intact step length; cluster 3 demonstrated moderate symmetry with the highest cadence. This study highlights that customized rehabilitation targeting specific gait patterns—such as strengthening muscles to increase single-limb support and step length, and modulating cadence—could enhance gait performance in individuals with uTFA.