Multi-Source Information Based Short-term Taxi Demand Prediction Using Deep-Learning Approaches

Abstract

Taxi service is of great importance in public transportation for its flexibility and convenience. As the advent of smartphones, traditional taxi service is under revolution. Instead of randomly taking a taxi, people are able to call for a taxi pickup by the smartphone applications (e. g. Uber and Lyft). Once the nearest taxi driver receives the request, he can head to the pick-up location. While in most urban areas, the taxi service demand and supply are imbalanced, which indicates that sometimes passengers wait too long for taxis and drivers roam too long with vacant taxicabs. Short-term demand prediction is of great importance to the on-demand ride-hailing services. Predicted taxi demand information can facilitate efficient operations and rebalance both demand and supply sides. This paper proposes a multi-source information based spatiotemporal neural network (MSI-STNN) deep learning architecture to predict short-term taxi demand. The model fuses pick-up and drop-off time-series data, weather information, and location popularity data, using three deep-learning models, including stacked convolutional long short-term memory (ConvLSTM) model, stacked long short-term memory (LSTM) model and convolutional neural network (CNN) model. ConvLSTM captures the spatiotemporal features of pick-up and drop-off time series. LSTM and CNN extracts information of weather and popularity. Case studies were performed to predict short-term pick-up demand at zonal levels in 15 minutes using New York taxi data. The experiment results validate the accuracy performance of the proposed approach comparing with state-of-art time-series and deep learning approaches, including ARIMA, ConvLSTM, CNN, and LSTM.