Forecasting public transit ridership amidst COVID-19: a machine learning approach

Transit demand prediction is critical for effective public transit planning and operations, particularly in the unpredictable environment caused by the COVID-19 pandemic. Since the beginning of 2020, governments worldwide have implemented various non-pharmaceutical interventions to control the spread of the pandemic, impacting multiple sectors, such as education, health, industries, agriculture, and transportation. However, the specific effects of these interventions on transit ridership have yet to be quantified, making it a challenging task. This study shows that machine learning can be used to develop a model that correlates the impact of imposed interventions with other relevant variables on transit ridership. Four prediction models based on LSTM were developed, each using different groups of input variables related to both ridership and the pandemic. These input groups include seasonal data, non-pharmaceutical interventions, and COVID-19 statistics. Despite the difference in input variables, all models aimed to predict daily transit ridership. The results showed that the proposed model accurately and reliably maps the complex relationship between interventions and transit ridership. The predicted and actual results align closely, as evidenced by a high coefficient of determination (R2) of 0.96, with data points clustered along the regression line. Implementing this model can be highly beneficial for managing transit demand during disasters and emergencies, where predicting transit demand is critical for operational and planning purposes. Graphical Abstract