Identifying the feature of bus passenger flow is the key to improve the quality of short-time prediction． However，due to equipment failure，data collection constraint and other reasons，the attributes of passenger flow data are often incomplete，which brings challenges to feature recognition and passenger flow prediction． The card data of No. 104 bus in Changsha，lacking passenger attribute data，was taken as the sample in this paper． The cor- relation between card number and travel time was used to identify passengers’travel frequency，which was used as a variable to distinguish the feature of travel． The total passenger flow set was divided into different feature subsets， and the optimal value of variable was determined according to the subset size and variance value，and the attributes of passenger flow were inferred． Compared with direct prediction of total passenger flow，the Seasonal Auto-Regres- sive Integrated Moving Average ( SARIMA) model established for each subset was respectively used for prediction． The out of sample Mean Absolute Error ( MAE) obtained by integrating is improved by 36. 11% ． The fitting degree of prediction model based on the feature of passengers’travel is 0. 95，thus can effectively identify the feature of bus passenger flow．