Taxi Demand Prediction Using LSTM and Optimized Taxi Geo-distribution

Abstract

AbstractLocation and time-based taxi demand prediction is an essential aspect for e-hailing taxi service providers to monitor the distribution of taxis around the city, which helps in maximizing utilization, minimizing average passenger waiting time, and average cab idle time and efficient fuel usage. Two critical challenges faced to achieve the above goal are an accurate and quick prediction of the demand and organizing the driver fleet according to the prediction. Hence, we propose a system that can be center indirectly managing the dispatch of taxis according to future demand. The system deploys a sequence learning model LSTM (time) and clustering algorithms (location) for predicting taxi demands in different areas (clusters) of a city. For organizing driver distribution, the system implements a rank-based mapping algorithm that minimizes the average vacant taxi travel of the drivers and maintains demand-supply balance. KeywordsSequence learning model LSTMRecurrent neural networkGeo-locationTime-based taxi demand predictionAverage vacant taxi travelDriver mappingRank tableBoxplot analysisDemand-supply matchHaversine distance