The Prediction Of Taxi Demand Based On Machine Learning

In recent years,with the continuous advancement and rapid development of big data technology and artificial intelligence,as well as the concept of smart city,the intelligent transportation system in our cities is undergoing profound changes.The transportation system will be the "smart traffic" which combines different advanced technologies,such as artificial intelligence and Internet+.As an important part of the city's public transportation system,taxis are able to provide the convenient and comfortable personalized services,and have become an indispensable trip mode in citizens' daily life.In view of the current contradiction between supply and demand related to the taxi service and passenger demand,it is of great significance to effectively predict the passenger demands in the coming time by relevant methods.This paper firstly pre-processes the taxi trip data in New York City,and analyzes the spatial and temporal distribution characteristics of taxi demands based on the data processing,as well as the influence of external factors,such as weather conditions.Secondly,for the prediction of taxi demands in a single region over a period of time,this paper selects the prediction methods related to machine learning,and uses multiple linear regression,support vector regression,random forest and Xgboost algorithm to predict the number of passengers at JFK airport respectively.Then the paper analyzes the predictive performance of different models.Finally,this paper adopts the deep learning methods to predict the taxi demands in multiple regions.Due to the shortcomings of the traditional long-term memory neural network and ConvLSTM network in dealing with the issues of spatio-temporal characteristics,this paper proposes a prediction model called CNN-Resnet-LSTM,and proves that this model outperforms other models in predicting the taxi demands under certain conditions,and analyses the factors and degradation models.