Taxi Destination Prediction with Deep Spatial-Temporal Features

Taxi destination prediction can grasp the flow direction of the taxi, facilitate the taxi dispatches. The existing methods mostly use the original features of the trajectory to predict and ignore the spatio-temporal features behind the original features, resulting in lack of spatio-temporal information of the trajectory. To address the above problem, we propose a taxi Destination Prediction method with Deep Spatial-Temporal features (DPDST). Firstly, we take advantage of sliding window to calculate the high-level features based on the speed and the turning rate. Secondly, we pass the auto-encoder to learn deep spatial-temporal features from high-level features of the trajectory. Finally, obtained deep spatial-temporal features and original features are combined as input to LSTM (Long Short-Term Memory Network)-based model. Experiments demonstrate that the accuracy is 9% and 7% higher than traditional RNN and LSTM model, and the average distance error is reduced by 1.3km and 1km respectively.