In this fast developing world, the number of motor vehicles is increasing rapidly but road resources remain limited, causing severe congestion problem of city traffic. In order to predict short-term traffic condition accurately, we propose a short-term traffic condition prediction method for urban road network based on improved support vector machine. As outliers inevitably exist in collected traffic data, our method can effectively deal with the negative effects of outliers, thus enhance the robustness of the model and improve the generalization ability. Experimental results show that the improved method has better classification accuracy than other machine learning methods, which further validates the effectiveness of our method.