A method of recommending waiting roads to passengers based on vacant taxi number prediction

Currently, the low operation efficiency of taxis is a common problem faced by urban taxi service systems. When there is a high vacancy rate of taxis, it can still be difficult to find a taxi, mainly due to the uninformed selection of waiting roads on the part of passengers. To solve this problem, this paper proposes a method of recommending waiting roads to passengers based on vacant taxi number prediction. By extracting the spatio-temporal and ambient features influencing the vacancy rate from the historical trajectory big data and weather data of urban taxis, this method first predicts the vacant taxi numbers of various roads using the CatBoost framework. Then the probability of taking a taxi in a future time slot on each road is calculated. Finally, the adjacent road with the best probability to the passenger based on his/her current location is recommended. Thus, the difficulty of taking a taxi is overcome to some extent. This study conducted a simulation experiment for model training and validation based on the historical trajectory big data of taxis in Xi’an and compared the proposed method with LightGBM and SVM. According to the experimental results, the proposed CatBoost method was superior to other methods in root-mean-square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) indices, and it has some practical values for resolving the taxi-passenger contradiction.