Prediction of Hourly Subway Ridership based on Artificial Intelligence Algorithms Using Weather Information

This paper presents a model for predicting hourly subway ridership based on weather conditions. The goal is to provide efficient subway system service based on prediction model results. Three artificial intelligence algorithms, Multiple Linear Regression (MLR), Random Forest Regression (RFR), and Multi-Layer Perceptron (MLP), were used to compare and analyze the performance of our prediction model for hourly subway ridership. The performance of the prediction models was optimized across two subway stations with diverse and different regional characteristics rather than a single station. To compare the prediction results, 8 out of 15 initially considered independent variables were selected and used for performance enhancement according to the correlation and feature importance with the dependent variable. The outcomes indicated that the subway station heavily used by office workers is almost unaffected by weather variables, while the subway station used mostly by people enjoying leisure activities is more affected by weather variables. Across all datasets, the model using the MLP algorithm outperformed the alternative models in accurately predicting subway ridership.