Improving Prediction for taxi demand by using Machine Learning

Abstract

Many problems and accidents are becoming increasingly occurring due to the increased number of vehicles on the streets. Therefore, much research has been submitted to help reduce vehicle problems such as accidents, congestion, and others, such as predicting taxi requests in the regions. Taxis are currently a high percentage of the street's number of vehicles, and if they are directed correctly to their target (passengers), this will contribute to reducing the congestion in the streets. Relying on developed technology such as Vehicular Social networks (VSN) can provide the necessary data for drivers to update their data continuously when there is a network connection. Some previous related works are criticized according to this task. This paper suggests improving taxi demand prediction in the regions based on data preprocessing. This study focuses on a comparison among four machine learning algorithms used for taxi request prediction and finding the best one in terms of execution time and error rates. Finally, Recent data was used for the first three months of 2021 and 2022, where 70% for training and 30% for testing for the year 2021, while the year 2022 is all data for testing. The results show that the Random Forest model outperforms LSTM, ANN, and linear regression in terms of error rates, and it obtained MSE 4.3 * 10−4 and RMSE 2.09 * 10−2.