WDA: Estimation of Travel Time for Taxi Using Wide and Deep Learning Together with Target Attention

Abstract

There is an increasing interest in ways to use data and machine learning methods to optimize the operation of transportation system. In this paper, the author focuses on estimation of time arrival (ETA) for taxi. As taxi is one of the popular transportation vehicles in the urban area, predicting the time of arrival for taxi would help the customer to better estimate time required for the trip, especially in case of urgencies. The author develops a comprehensive neural network with both wide and deep components, and attention component for this predictive job. The author also evaluates the network with huge historic data and compare it with industry existing solutions. This network model is proven to have a better prediction accuracy under extreme traffic and weather condition with reasonable shorter training time.