Identifying Linkages between Demographics, Behaviors, and Road Accident Frequency - A Machine Learning Approach in England

Aug 5, 2026

This study addresses the difficulty of collecting comprehensive, driver-level behavioral data that is linked to more than 70% of road accidents. It examines demographic factors and their indirect relationship with risky driving by synthesizing existing literature that connects demographics to unsafe behaviors. Regression-based machine learning models are applied across all of England to relate behavioral features, demographic characteristics, and accident frequency. Results match prior research and offer a practical methodology to investigate behavior-accident links at a broader scale under data constraints. The approach supports estimating accident rates from readily available demographic information.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages12

Reading Time30 min

AccessFree to download

Introduction

Human Behavior, Demographics, and Road Accidents

IDENTIFYING LINKAGES BETWEEN DEMOGRAPHICS, BEHAVIORS, AND ROAD ACCIDENT FREQUENCY: A MACHINE LEARNING APPROACH IN ENGLAND (1108) Dakota Aaron McCarty 1 1 Department of Urban Policy & Administration, Incheon National University, Yeonsu-gu, Incheon, Korea Abstract. This study addresses the challenge of collecting comprehensive data on individual drivers' behavior, which has been found to contribute to over 70% of road accidents. It focuses on demographic factors and their indirect impact on accident rates by leveraging existing literature on the connection between demographics and risky driving behaviors. A review of literature identifies demographic characteristics correlated with risky driving behaviors. Using regression-based machine learning models, the investigation covers all of England, UK, aiming to establish connections between driving behaviors, demographics, and accident frequency. The study's results align with previous findings and provide a valuable methodology to investigate behavior-accident links on abroader scale despite data limitations. The findings support utilizing readily available demographic information to estimate road accident rates. Keywords: Driving behavior; Traffic accidents; Cohort analysis; SHAP values; Extra Trees Regressor. 1. Introduction The study of human behavior presents inherent complexities within the field of behavioral sciences, primarily due to the challenges associated with measuring behavior across diverse populations and timeframes. Comprehensive research in this area necessitates a thorough understanding of various influences and relationships, aiming to control for a wide range of external factors (Kelly and Barker, 2016) . While extensive research exists on other behavioral issues, such as substance abuse and eating habits, there has been relatively less attention given to risky driving behavior. This is a notable issue as driving behavior is dissimilar from many other potentially risky behaviors. For instance, modifying behavior related to ‘rare events’ like road accidents poses challenges since individuals engaging in such behavior often do not experience immediate negative consequences in their daily lives. Additionally, habits—both positive and negative—such as exercising or smoking, gradually develop and become ingrained in daily routines, whereas the need to change driving behavior is typically prompted by accidents or reflexive responses (Kelly and Barker, 2016) . When studying the general population, further challenges arise when attempting to control for factors like exposure and experience (Brown, 1982) . These complications are particularly relevant when attempting to investigate the impacts of risky driving behaviors on road safety. To overcome some of these limitations, researchers often narrow the scope of their investigations by focusing on specific demographics, geographic areas, or other controllable factors that facilitate data collection (e.g., through surveys or in-car monitoring devices). However, this narrowing of focus inadvertently reduces the generalisability of the research to new areas and populations. Consequently, there is a need to explore new variables that can serve as proxies to estimate driver behavior and the risk of road accidents, as highlighted in previous studies. Thus, the primary objective of this research is to identify potential proxies for estimating risky behaviors that can be utilised in future road accident prediction models. Numerous studies have established the significant role of human behavior in causing road traffic accidents (Andrew Yockey and Barroso, 2023; Bener et al., 2009) . Moreover, an expanding body of research supports the effectiveness of behavioral science approaches in identifying and mitigating accident-prone behaviors (Gielen and Sleet, 2003) . This study aims to explore readily obtainable features, mitigating a limitation in the field around data collection and availability, which have been found

FAQs

What problem does this research aim to solve about road accidents and data collection?

It targets the challenge of obtaining comprehensive data on individual drivers’ behaviors, which are associated with the majority of road accidents.

How does the study connect demographics to road accident frequency?

It reviews literature to identify demographic characteristics linked to risky driving behaviors, then uses regression-based machine learning to relate these features to accident frequency.

What modeling approach and outputs are used to support the findings?

The study applies regression-based machine learning models and uses tools such as SHAP values, including an Extra Trees Regressor, to investigate relationships between features and outcomes.

Aug 5, 2026

Upload by Valentina