Public Transport Passenger Count Forecasting in Pandemic Scenarios Using Regression Tsetlin Machine. Case Study of Agder, Norway

Challenged by the effects of the COVID-19 pandemic, public transport is suffering from low ridership and staggering economic losses. One of the factors which triggered such losses was the lack of preparedness among governments and public transport providers. The losses can be minimized if the passenger count can be predicted with a higher accuracy and the public transport provision adapted to the demand in real time. The present paper explores the use of a novel machine learning algorithm, namely Regression Tsetlin Machine, in using historical passenger transport data from the current COVID-19 pandemic and pre-pandemic period, combined with a calendar of pandemic-related events (e.g. daily number of new cases and deaths, restrictive measures for pandemic containment), to forecast public transport patronage variations in a pandemic scenario. Results show that the Regression Tsetlin Machine has the best accuracy of forecasts when compared to four other models usually employed in the public transport forecasting field. We also observed variations of the prediction accuracy in relation to the period of the pandemic in which the trained models are applied. The underlying reasons for the relative passenger count variations are also examined using the properties of the Tsetlin Machine.