Abstract
The complexity of predicting hotspot areas for taxicab services has recently increased with the popularity of ride-hailing services such as Uber and Lyft. In this paper, we first reveal that passengers in certain areas prefer ride-hailing services over taxicab services at certain times, and propose an enhanced LSTM model for predicting the hotspots areas of ride-hailing and taxicab services. It learns the spatio-temporal aspects of ride-hailing and taxicab services' pickup and drop-off orders to predict future orders every 10-minutes, based on which, a hotspot rec-ommender system that comprises a pipeline of recommending the best hotspot areas to taxicab drivers by taking both regional passenger's preference and the distance of the driver's current location into account. The evaluation of our approach on real datasets of five ride-hailing and taxi-cab service providers in New York City (NYC) demonstrated sensible accuracy which is comparable to baseline models. We demonstrate the benefits of our hotspot recommender algorithm over two scenarios considering the NYC dataset and our demand and supply prediction model in terms of suggesting the best hotspots taxicab drivers should target.