We first investigated the statistical characteristics of the 8 features, computing the mean, standard deviation and ROC areas for the individual features. As seen from Table 1, the means for velocity, cadence, stride length and stride time do not differ, suggesting that they are similar for both classes and therefore have poor discriminative power. This is confirmed further by individual ROC values in the range of 0.524–0.589 indicating the two classes are not linearly separable using these features alone. The symmetry indexes of step length step time, single support and double support features have significantly higher ROC values, that is larger than 0.65. This suggests that the asymmetry index of the temporal measures: single support and double support is greater in patients with OA, since these patients spend a greater portion of the gait cycle on the non-operated leg. It was also found that the values of cadence were much larger in magnitude than the other features. Thus all features were normalized to zero mean and unit standard deviation to facilitate the training of SVMs