Prediction of Travel Purpose Based on the Long Short-Term Memory Network

GPS can record most travel trajectories, which helps predict travel demand. However, the raw data does not contain travel purpose and thus cannot fully replace a traditional resident travel survey. There are a lot of studies on the prediction of travel purpose, but most of them predict the travel purpose with a single travel without considering the sequence of travels for individuals. To fill this gap, this study aims to develop a long short-term memory network (LSTM) model by combining individual travel sequences with points of interest (POI) data from Baidu Map to predict travel purpose. The model’s performance is evaluated on Chengdu’s resident travel survey data in 2016 and compared to random forest (RF). The influence of different input features on the model is also analyzed. It turns out that using POI data and considering individual travel sequence can improve prediction accuracy for travel purpose significantly.