Predictive modeling study demonstrates robust port traffic forecasting using calendar conditioning, indicating regime specialization requires stable model dominance across time.
Key Points
To assess whether temporal heterogeneity in port access traffic is better addressed through a unified calendar-conditioned forecasting ensemble or fixed regime-based model specialization.
Forecast daily car and freight-truck arrivals at an urban non-containerized port across 1-, 7-, and 14-day horizons using statistical, machine-learning, and composite models.
Implemented a leakage-free rolling-origin protocol to compare a calendar-conditioned ensemble against prespecified regime-based assignments across distinct development, 2018 test, and 2019 temporal stress-test periods.
The unified ensemble achieved next-day R2 values of 0.75 for trucks and 0.80 for cars on the 2018 test period, maintaining 0.73 and 0.75 during the 2019 temporal stress-test.
Although fixed regime assignment improved car forecast accuracy during the 2018 test period, model-regime dominance lacked sufficient stability across forecast horizons and evaluation periods to justify specialized model switching.