Reliable and performance short-term traffic flow prediction are critical to providing effective traffic control and trip planning. Nonlinearity and the complexity of the transportation system continue to be formidable obstacles to traffic flow prediction. In this paper, we employ long short-term memory (LSTM) and bidirectional long short-term memory (BILSTM) to investigate the impact of heterogenous data input configuration on performance prediction. Traffic flow (5 vehicle types), environmental conditions (10 conditions), and events (3 events) are used as input to predict traffic flow. The experimental results indicate that the model with heterogeneous data can enhance the prediction model’s performance.