Examining Seasonal Changes in Light-Vehicle Traffic Volume on Freeways Under Extreme Weather Conditions: A Combination of Temporal Statistical and Data Mining Non-Parametric Techniques

Abstract

Today, with the increasing changes in weather patterns and the huge amount of data related to weather and traffic in different parts of the freeway network, the use of data mining methods to quantify the impact of weather on traffic flow is inevitable. The main objective of this study is to present a geostatistical method for computing and analyzing the effects of base and extreme cases of weather variables on light-vehicle traffic volumes on freeways, with an emphasis on temporal changes on different days of the week and between daytime and nighttime. In the proposed method for statistical analysis, the parametric test of two-way analysis of variance was used. In the following, with the development of a nonparametric method based on the classification and regression tree (CART) decision tree algorithm, the weather-related parameters with the greatest effect on traffic volumes were investigated separately for different seasons. For this purpose, nine years of statistics covering traffic and weather data for the studied freeway were analyzed. The computational results show that a fall in temperature of more than 35% on weekdays during the cold season and the parameters of horizontal visibility and rainfall during the day and temperature during the night in the spring cause a reduction of traffic volume of more than 45%. This study has shown that the combination of data-driven parametric and nonparametric data mining techniques is effective for traffic managers in planning and traffic control under extreme weather conditions. In this regard, an adverse weather conditions dynamic message sign (AWCDMS) framework was proposed as a practical way to warn drivers of adverse weather.