Predicting Requests in Large-Scale Online P2P Ridesharing

TL;DR

This paper investigates the impact of request prediction on large-scale P2P ridesharing efficiency, showing potential benefits with perfect forecasts and analyzing neural network prediction capabilities.

Contribution

It assesses the benefits of request prediction in P2P ridesharing and compares neural network models with baseline predictors.

Findings

Perfect predictor improves total reward by 5.27%

LSTM neural network does not outperform baseline in request prediction

Forecast horizon of 1 minute is considered

Abstract

Peer-to-peer ridesharing (P2P-RS) enables people to arrange one-time rides with their own private cars, without the involvement of professional drivers. It is a prominent collective intelligence application producing significant benefits both for individuals (reduced costs) and for the entire community (reduced pollution and traffic), as we showed in a recent publication where we proposed an online approximate solution algorithm for large-scale P2P-RS. In this paper we tackle the fundamental question of assessing the benefit of predicting ridesharing requests in the context of P2P-RS optimisation. Results on a public real-world show that, by employing a perfect predictor, the total reward can be improved by 5.27% with a forecast horizon of 1 minute. On the other hand, a vanilla long short-term memory neural network cannot improve upon a baseline predictor that simply replicates the previous day's requests, whilst achieving an almost-double accuracy.