Abstract: Because stock market volatility is of mutability and variability,and the distribution of the time series data does not follow the normal distribution,the traditional time series forecast algorithms are difficult to accurate prediction．The stock market tracking prediction algorithm based on Stream Feature Model was proposed (SFM-PG).It is builds the Bayesian networks based on correlation between stocks,selects the Markov Blanket of the target stock as its peer group,and gives a windows tracking prediction model based on the proximity between peer group,through dynamiclly updating the weight of peer group to tracking prediction,effective avoids the influence of the non-normal distribution of time series data on prediction．And then,using the sliding window to extract the feature of the time series data to formation stream feature,and extracting the stream feature model by matching with knowledge base of base,using the knowledge of stream feature model to adjust the predicted results,in order to reduce the prediction error introduced by mutability．Finally,the practicability and effectiveness are showed in the experiment on the network of plate of the Shanghai stock.

Key words: Stream feature,Stream feature model,Peer group analysis,Stock market price forecasting