Predicting high taxi demand regions using social media check-ins

Abstract

The supply and demand of taxis in a city sometimes can be highly imbalanced. In some regions, passengers are unable to get rides even after a long wait, while in other regions many taxis are loitering without any passengers. Therefore, an intelligent taxi demand forecasting system is desired to not only allow the taxi drivers to meet passengers needs, but also reduce their futile roaming. In this paper, we attempt to forecast the high taxi demand regions in NYC by using social media check-ins. Our approach includes four primary steps. First, we locate the popular venues in NYC and then cluster these venues into high geo-resolution regions that cover most of the popular areas in NYC. Second, we further refine these regions into functional zones via their internal venue distributions. Third, we inspect how temporal and geo-spatial factors impact the taxi demand in different zone types. Finally, we try several ways to improve the forecast accuracy, and our best way achieves 82% validating accuracy and 80% testing accuracy.