Abstract:
Transit demand prediction is critical for the effective planning of public transit
operations, particularly in the very unpredictable environment caused by the COVID-
19 pandemic. Since the beginning of 2020, many governments worldwide have
implemented various non-pharmaceutical interventions to control the propagation of
the pandemic. These interventions have disrupted daily routines and severely impacted
multiple sectors of life, including education, health, economy, tourism, industries,
agriculture, employment, and, most notably, the transport sector. However, the impacts
of these interventions on transit ridership have not been measured or quantified, making
it a challenging task. This study shows that machine learning may be used to develop a
model that correlates the impact of imposed interventions with other relevant variables
on transit ridership. An LSTM based four prediction models were developed, each of
which using different variables input groups, that are related to both ridership and the
pandemic. The three input groups are seasonal data, non-pharmaceutical interventions,
and COVID-19 statistics data. Although these models differ in their input variables,
they are all built to predict the same output, which is the daily transit ridership. The
findings suggest that the proposed model is able to map the complex relationship with
a high degree of accuracy and reliability. The predicted results and the actual results are
fairly comparable. With a coefficient of determination (R2) of 0.96, they are clustered
along the regression line. The application of this model will be highly beneficial for
managing transit demand during disasters and emergency situations, where the
prediction of transit demand is critical for operational and planning purposes.
