A Universal Automated Data-Driven Modeling Framework for Truck Traffic Volume Prediction

Mahdavian, Amirsaman; Shojaei, Alireza; Salem, Milad; Laman, Haluk; Eluru, Naveen; Oloufa, Amr A. · 2021 · Crossref

DOI: 10.1109/access.2021.3099029

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the critical need for accurate truck traffic volume prediction to support highway planning, pavement design, and congestion management. Motivated by the substantial increase in freight travel due to urbanization and economic growth, the authors identify a gap in existing literature: the lack of a universal, automated framework that incorporates inclusive predictors, broad datasets, and robust cross-validation using both linear and non-linear algorithms. Previous models were often limited to specific case studies or selective variable subsets, lacking generalizability and automation. To fill this gap, the study develops a universal automated data-driven modeling framework. The methodology utilizes historical monthly average daily truck traffic (MADTT) data from 259 sites across six Florida interstate highways, spanning from 2001 to 2017. The dataset comprises 52,836 data points and includes 59 candidate predictor variables categorized into construction market, energy market, socioeconomic, U.S. economy, road characteristics, temporal, and spatial groups. The framework employs a hyperparameter optimization pipeline (grid search) to automatically select the optimal feature selection method and modeling approach. It tests five linear and four non-linear machine learning algorithms, including Random Forest, Support Vector Regression, and Neural Networks, using the Scikit-learn library in Python. Data preprocessing involved standardization and time-series-appropriate partitioning into training, validation, and test sets to preserve temporal continuity. The results demonstrate the superiority of non-linear models over linear models in generalizing and predicting truck traffic volumes. Specifically, the Random Forest algorithm achieved the highest performance, predicting truck traffic with 86% accuracy on the test dataset. The analysis identified spatial variables as the most significant predictor group, followed by road characteristics. The automated framework successfully minimized Mean Absolute Percentage Error (MAPE) by selecting the best-performing model and feature subset for the given data. The significance of this work lies in providing a customizable, labor-efficient tool for transportation planners and decision-makers. Unlike previous case-specific models, this universal framework can be adapted to local datasets and scenarios, allowing users without extensive data analysis expertise to generate accurate long-term truck traffic forecasts. By automating the selection of features and algorithms, the study offers a robust solution for improving the reliability of freight movement predictions, thereby aiding in infrastructure investment policies and environmental impact analyses.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• traffic density