In this paper, we proposed a hybrid system of two powerful machine learning schemes, namely the generative model HMM (hidden Markov model) and the discriminative model SVM (support vector machines), which can concertedly enhance the accuracy of human action recognition. The basic idea of our hybrid method is that the temporal characteristics of the sequence data are modeled by HMM state transitions and both the SVM model and Platt’s method obtained through the preprocessing are used to compute the emission probabilities of observations given at each state in HMM. In order to evaluate the performance of the proposed model, we have conducted an experiment that is able to compare our hybrid model with other classical methods using a KTH dataset. From our experimental results, we note that our hybrid approach shows better or poor performance than existing methods in various fields of human action.