Forecasting Daily MRT Passenger Flow in Taipei Based on Google Search Queries

In recent decades, the advancement of the transportation system raises people's reliance on public transportation. Taipei metro is one of the most often-used modern transportation in Taiwan with around 2 million ridership per day. Forecasting passenger flow has been a critical topic in transportation research, with the purpose to facilitate traffic control. Accurate prediction on passenger flow would be essential for the efficiency in terms of operation and management of urban metro systems. In this paper, we focus on forecasting the daily Taipei metro passenger flow with various external covariates including Google search quires, calendar variables, and disaster variables to examine their forecasting power. Four existing methods such as linear regression, least absolute shrinkage, selection operator (LASSO), random forest, and autoregressive integrated moving average (ARIMA) are employed. The result suggests that random forest appears to yield the most competitive result among models and including Google search queries as covariates can in fact improve the accuracy of prediction.