The Effectiveness of Socio-Economic Context in Short-Term Highway Traffic Forecasting
Ohsung Kwon1, Kyoung-Soub Lee2,*
1 Department of Business Intelligence, Ajou University, Suwon, Republic of Korea
2 School of Mobility Engineering, University of Ulsan, Ulsan, Republic of Korea
* Corresponding Author: Kyoung-Soub Lee. Email: 
(This article belongs to the Special Issue: Advances in Time Series Analysis, Modelling and Forecasting)
Computers, Materials & Continua https://doi.org/10.32604/cmc.2026.083882
Received 12 April 2026; Accepted 18 August 2026; Published online 08 September 2026
Abstract
Traffic congestion incurs massive social and economic costs. Accordingly, the importance of transportation management and planning is gradually expanding. In response, the establishment of intelligent transportation systems utilizing advanced technology is increasing, as are global attempts to predict traffic. However, predicting traffic volume time series is challenging because they are influenced by various environmental, social, and economic factors, and thus exhibit high nonlinearity and stochasticity. Accurate prediction of highway traffic, which forms the backbone of a nation’s transportation and logistics, is crucial. Therefore, this study examined whether meteorological, environmental, and economic covariates improve nationwide highway traffic forecasts in South Korea beyond what calendar information alone provides. Our dataset consisted of daily traffic volumes aggregated over all expressways, together with calendar variables and additional factors such as temperature, precipitation, particulate matter concentrations, and fuel prices. Using these inputs, we compared twelve models spanning naive, statistical, ensemble machine learning, recurrent, convolutional, and Transformer families. Across these models, all evaluated on an identical test period, calendar variables accounted for essentially all of the achievable accuracy, reducing the error of the best model from 4.40% to 3.67% mean absolute percentage error (MAPE), whereas single-station metropolitan weather, air-quality, and fuel-price series yielded no measurable improvement. Among the evaluated models, the temporal convolutional network (TCN) achieved the highest accuracy. This result indicates that, for daily policy-oriented forecasting, the choice of calendar inputs and an appropriate level of model complexity matter more than architectural sophistication alone.
Keywords
Time series forecasting; deep sequence modeling; temporal convolutional network; holiday effects; ablation analysis; nationwide demand forecasting
Open Access
Login
Register
Submit a Paper
Propose a Special lssue
Download PDF
Downloads
Citation Tools
