Machine Learning Based Taxi Demand Prediction

Abstract

Traffic is the pulse of any city that impacts the social, economic and healthy life of millions of people. Demand prediction, mainly taxi demand prediction, plays an important role in intelligent transportation systems in smart cities. The taxi demand prediction problem is to precisely predict the number of taxi requests for a certain region at a certain time-bin based on the information provided by previous taxi requests. In this research, we used machine learning methods, such as random forest regression RFR, support vector regression SVR, and artificial neural network ANN, to accurately predict taxi demand. Our goal is to improve the accuracy of taxi demand prediction by using different ways of prediction. Therefore, we studied the effect of the location on taxi demand prediction. We consider the location of the pickup as a cluster, three decimal rounded coordinates or neighborhood. The research uses yellow-taxi requests in New York City NYC as a dataset. We refined our data by adding weather information. We found that location features affect the prediction result and the performance of the model. NYC neighborhoods as a location feature are the best in the model’s pervi formance, as in the NN model. We noticed the importance of the weather features with the three decimal rounded location coordinates. The NN model approximately gives the best performance