Region-level Ride-hailing Demand Prediction with Deep Learning

Abstract

The exponential growth of ride-hailing demand has disrupted the urban transportation and changing the way people travel. To alleviate the shortage problem, many researchers have done numerous experiments to predict the region-level ride-hailing demand which can arrange the vehicles more efficiently. In this paper, we use the online ride-hailing order dataset of Haikou, Hainan Province, China, released by Didi Chuxing. We apply historical models, MLP, CNN and ConvLSTM model for region-level ride-hailing demand prediction in the next hour. Compared to the traditional deep learning models, ConvLSTM has a better performance with the lowest RMSE (4.796) when the input historical length is 24 hours.