With the development of anonymous communication technology, it has led to the fact that the network monitoring is becoming more and more difficult. If the anonymous traffic can be effectively identified, the abuse of such technology can be prevented. Since the study of machine learning is rapidly developing these years, this paper applies the Random Forest Algorithm – a semi-supervised learning method – into the traffic detection of Shadowsocks. We can get over 85% detection accuracy rate in our experiments after applying Random Forest Algorithm by collecting train set, gathering features, training models and predicting results. With the scale of train set and test set increase, the detection accuracy rate gradually increases until it becomes constant. We will make several adjustments on train set, test set and feature set to reduce the false alarm rate and false rate when detecting.