An ensemble learning approach for the kaggle taxi travel time prediction challenge

T. Hoch. An ensemble learning approach for the kaggle taxi travel time prediction challenge. volume 1526, pages http://ceur-ws.org/Vol-1526/paper22.pdf, 1, 2016.

Autoren
  • Thomas Hoch
Editoren
  • A. Martínez-Usó
  • J. Mendes-Moreira
  • L. Moreira-Matias
  • M. Kull
  • N. Lachiche
BuchProceedings of the ECML/PKDD 2015 Discovery Challenges
TypIn Konferenzband
VerlagCEUR
SerieCEUR Workshop Proceedings
Band1526
Monat1
Jahr2016
Seitenhttp://ceur-ws.org/Vol-1526/paper22.pdf
Abstract

This paper describes the winning solution to the Taxi Trip Time Prediction Challenge run by Kaggle.com. The goal of the competition was to build a predictive framework that is able to predict the final destination and the total traveling time of taxi rides based on their (initial) partial trajectories. The available data consists of all taxi trips of 442 taxis running in the city of Porto within one year. The presented solution consists of an ensemble of expert models combined with a spatial clustering approach. The base classifiers consist of Random Forest Regressors where as the expert models for each test trip where based on a combination of gradient boosting and random forest. The paper shows how these models can be combined in order to generate accurate predictions of the remaining traveling time of a taxi.