[Paper Review] Road Accidents in the UK (Analysis and Visualization)

Anjul Tyagi, Ayush Kumar+2 more|arXiv (Cornell University)|Jul 29, 2019

Traffic and Road Safety1 references4 citations

TL;DR

This study applies Multiple Correspondence Analysis (MCA) to visualize and correlate key factors in UK road accidents—postcode, day of the week, and driver age group—revealing patterns such as Liverpool (L) having peak accidents on Saturdays with drivers aged 26–35. It further uses hypothesis testing and time series analysis for variables not well-represented by MCA, finding a declining trend in annual accidents and significant gender-based disparities in accident frequency.

ABSTRACT

Analysis of road accidents is crucial to understand the factors involved and their impact. Accidents usually involve multiple variables like time, weather conditions, age of driver, etc. and hence it is challenging to analyze the data. To solve this problem, we use Multiple Correspondence Analysis (MCA) to first, filter out the most number of variables which can be visualized effectively in two dimensions and then study the correlations among these variables in a two dimensional scatter plot. Other variables, for which MCA cannot capture ample variance in the projected dimensions, we use hypothesis testing and time series analysis for the study.

Motivation & Objective

• To identify and visualize the most influential variables in UK road accident data using dimensionality reduction.
• To understand correlations between accident location (postcode), day of the week, and driver age group through MCA.
• To analyze less visualizable variables—such as vehicle type, weather, and sex—using hypothesis testing and time series analysis.
• To predict trends in accident frequency over time and assess the impact of major events like the London Summer Olympics.

Proposed method

• Employed Multiple Correspondence Analysis (MCA) to reduce high-dimensional accident data into two dimensions for visualization and correlation analysis.
• Used a discrimination measure to assess how accurately each variable is represented in the MCA projection, prioritizing postcode, day of the week, and driver age group.
• Applied Welch’s t-test for hypothesis testing to compare accident frequencies across groups (e.g., male vs. female drivers, summer vs. winter).
• Used autoregression for time series analysis to model and predict monthly accident trends from 2005 to 2014.
• Combined MCA with supplementary analysis for variables with low MCA representation, ensuring comprehensive insight extraction.
• Validated predictions using root mean square error (RMSE), which was 699.84 for the autoregression model.

Experimental results

Research questions

• RQ1 Which combinations of postcode, day of the week, and driver age group show the highest accident frequency, and how are they correlated?
• RQ2 Are there significant differences in accident frequency between male and female drivers, and how do they compare?
• RQ3 Does the occurrence of the London Summer Olympics in 2012 significantly affect daily accident rates?
• RQ4 Are areas near subway stations associated with higher accident rates compared to other regions?
• RQ5 What is the long-term trend in UK monthly accident frequency from 2005 to 2014, and can it be reliably predicted?

Key findings

• Liverpool (L) postcode experiences the highest number of accidents on Saturdays, primarily involving drivers aged 26–35 years.
• The number of daily accidents involving young drivers (18–25 years) is 85 to 89 higher than for older drivers (65–85 years).
• Male drivers are responsible for 428 to 439 more daily accidents than female drivers.
• Areas near subway stations have 9 to 29 more daily accidents than non-subway areas.
• The number of daily accidents in summer is 15 to 30 higher than in winter.
• The autoregression model predicted monthly accident trends with a root mean square error of 699.84, indicating a declining overall trend from 2005 to 2014.