On analyzing and predicting regional taxicab service rate from trajectory data

Abstract

Taxicab companies want a solution for undersupply (oversupply) problem to boost profits. Finding regional taxicab demand is the key for reducing this disequilibrium. In this paper we investigate a taxicab demand model characterized by estimating demand distribution and recovering sparse data. When more and more trajectories accumulate, statistical characters gradually emerge, revealing a spatiotemporal correlated model. Three methods are addressed on this model: Parzen window estimation is used to get every-hour TSR (taxi service rate). Then, we leverage collaborative filtering to recover corrupted data. A TSR based neural network is to predict the demand. Experimental study is on real Beijing trajectory data, the result demonstrates that our proposed methods are able to feature taxicab demand and to provide dynamic demand prediction.