Introduction: Subject-dependent responses to clinical interventions have been considered as a major challenge in prescribing and predicting the effect of these interventions. It has been suggested that there may be groups of subjects who produce similar reactions to a specific intervention. These groups have been called functional groups. This study aimed to determining functional group susing lower limbs energy during running. Methodology: Mechanical energy of pelvic, thigh, leg and right foot of 118 students (58 males and 60 females) were calculated in stance phase of running. Principal component analysis (PCA) done for mechanical energy data and principal components which accounted for 95% variance of data was used as an input for Self-Organizing Map (SOM) with 40 neurons. Size of output matrices from SOM was 118×1600 that again was reduced dimensionality by PCA and the optimal group number determine using K-means clustering method. The classification rates were determined by a leave-one-out cross-validation technique that was applied using the support vector machine (SVM). Results: The k-means clustering algorithm was conducted with differentgroup numbers (k=5-8). The k-means clustering outcomecan slightly differ when executing for several times, so this algorithm repeated 10 times for each group number and the best result choose as a number of functional groups. Highest classification rate was 95.80% with five separate groups. Discussion: According to our results, the functional groups were well recognized with use of dimension reduction and unsupervised clustering methods. Most frequent classification methods are relying on selective features such as age and gender that had not desirable results. Number of groups and feature of functional groups is not well understood, thus determination of these groups using unsupervised clustering methods are very important.
