Outbound Time Prediction of Subway Passengers Based on Deep Forest Algorithm

To alleviate the congestion at subway gates and reduce the outbound delay for passengers, a prediction model of subway passengers outbound time based on deep forest (DF) regression was established. First, the definition of the outbound service process and outbound time of passengers in the subway station was clarified, and 15 factors that affect the outbound time of passengers were determined from the aspects of the passengers’ individual behavior and external environment. Then, select the optimal feature combination by combining Spearman coefficient and Genetic Algorithm, and the selected features are incorporated into DF model to predict the passenger departure time. The hyperparameters in the model were optimized by Tree Parzen Estimator (TPE), Covariance Matrix Adaptation Evolution Strategy, and random search methods for parameter optimization. Finally, through case analysis and comparative experiment verification, the results show that the DF prediction model optimized by TPE has the best prediction performance.