Accumulating evidence from human imaging data supports association between connectivity and outcome after stroke. However, whole-brain functional-connectivity (FC) involves high-dimensional data, which essentially calls for multivariate analysis. Regardless, univariate analysis has been dominantly employed for such studies. More insights into stroke recovery by employing machine learning techniques can be offered due to their multivariate capabilities. In this study, we investigated if residual sensory function (Tactile discrimination threshold, TDT score) of stroke patients can be predicted from resting-state FC. Our results show that TDT scores can be predicted more accurately by combining both low-order and high-order FC than low-order functional connectivity alone.