Forecasting Daily Changes in Passenger Flows in a Large Urban Agglomeration Based on Mobile Operator Data

Abstract

Purpose: to develop and comparatively analyze models for short-term forecasting of total daily intermunicipal passenger flows in a large urban agglomeration based on mobile operator data and to identify the approaches providing the highest forecast accuracy. Methods: linear, log-linear, and Poisson regressions, the Holt — Winters model, SARIMA, and hybrid models combining a regression component with timeseries models. Forecast accuracy was assessed using MAE, R2, and sMAPE. Results: it was found that the dynamics of total intermunicipal passenger flows are determined by a combination of calendar and weather factors, as well as pronounced weekly seasonality and short-term inertia. Models calibrated separately for each origin–destination pair were shown to provide higher forecast accuracy than models estimated on pooled panel data. The best results were obtained for the hybrid models “linear regression + HW” and “linear regression + SARIMA”. It was also established that excessive detailing of calendar factors does not improve forecast accuracy. Practical significance: the results obtained can be used in solving problems of operational management of an agglomeration transport system, including the assessment of the expected load on intermunicipal connections and the substantiation of transport service parameters. Forecasting total passenger flows based on mobile operator data makes it possible to complement forecasts for individual transport modes with an assessment of aggregate travel demand. This is of particular interest for the analysis and planning of suburban rail transportation.