A Statistical Model for the Prediction of the Taxi Trip Time in the City of Chicago

Abstract

This thesis addresses the problem of prediction of taxi trip duration for any given day, time, pickup point and dropo point. Data on taxi trips from the Chicago Data Portal is used. The main idea of the model is to cluster similar trips together and use the mean duration of all those clustered taxi trips to predict the duration of a new taxi trip in that cluster. Furthermore, for a possible additional reduction of prediction error, estimators from dierent days which are not signicantly dierent from each other are pooled together. It is shown that this procedure improves prediction error.