PROJECT TITLE :

Passenger Demand Prediction with Cellular Footprints

ABSTRACT:

An accurate forecast of the demand for passengers across the entire city enables providers of online car-hailing services to more efficiently schedule driver supplies. Previous studies either rely solely on passenger order history, in which case they are unable to capture the intricate interdependence of passenger demand, or they are restricted to grid region partitions, in which case they lose physical context. Recent advancements in mobile traffic analysis have helped to further our comprehension of how cities operate. In this piece, we present FlowFlexDP, a demand prediction model that applies to flexible region partitioning and incorporates regional crowd flow. The analysis of a cellular dataset that included 1.5 million users from a major city in China revealed a strong correlation between the demand for passengers and the flow of crowds. FlowFlexDP is able to extract order history as well as crowd flow from cellular data, and it uses a Graph Convolutional Neural Network in order to adapt its prediction for regions of a city that have arbitrary shapes and sizes. Evaluation on a large-scale data set of six online car-hailing applications using cellular data reveals that FlowFlexDP accurately predicts passenger demand and outperforms other methods that are considered to be state-of-the-art in terms of demand prediction.