Traffic Volume Prediction Using Data Mining Techniques on Historical Traffic and Weather Data

Gustavo Garcia-Vargas

Co-Presenters: Shea Hoff

College: Hennings College of Science Mathematics and Technology

Major: BS.COMPUTER/SCI

Faculty Research Mentor: Huang, Ching-Yu  

Abstract:

Traffic congestion poses significant challenges to urban transportation systems, impacting travel efficiency, fuel consumption, and environmental sustainability. This project focuses on predicting traffic volume using data mining techniques applied to historical traffic and weather data. The dataset used in this study is the Metro Interstate Traffic Volume dataset obtained from Kaggle, which contains 48,204 hourly traffic records collected between 2012 and 2018 along westbound Interstate 94 in the Minneapolis–St. Paul metropolitan area. The multivariate time-series dataset includes attributes such as temperature, rainfall, snowfall, cloud coverage, weather conditions, holidays, and timestamp informationThe research follows a systematic data mining process including data cleaning, transformation, and exploratory data analysis. Temporal features are derived from timestamp data, and correlation analysis is performed to examine relationships between traffic volume and environmental factors. Supervised learning methods are employed to model traffic volume as the target variable using historical and weather-based predictors. Model performance is evaluated using standard regression metrics to assess prediction accuracy.The hypothesis proposes that historical traffic and environmental attributes can effectively predict future traffic volume. Findings indicate strong correlations between traffic patterns, weather conditions, and peak versus off-peak driving hours. These results enhance the understanding of urban traffic behavior and contribute to the development of accurate traffic prediction models for smarter and more sustainable transportation systems.