Analysis of Run-Off-Road Accidents by Association Rule Mining and Geographic Information System Techniques on Imbalanced Datasets

Feifeng Jiang, Richard K.k. Yuen, Eric Wai Ming Lee, Jun MaSustainability 2020 · DOI

TL;DR

This paper introduces a novel framework combining Association Rule Mining (ARM) and Geographic Information Systems (GIS) to analyze the complex factors contributing to run-off-road (ROR) accidents, particularly fatal ones. By addressing the significant imbalance in accident data (far fewer fatal than non-fatal accidents) with a unique bootstrap-resampling method, the study uncovers key individual and interactive risk factors and maps their spatial distribution, offering crucial insights for improving road safety.

What Problem Does This Paper Address?

Road traffic accidents are a major global concern, with run-off-road (ROR) accidents being particularly deadly. Identifying the factors that lead to these accidents, especially fatal ones, is crucial for developing effective safety measures. However, existing research faces two main challenges:

  1. Imbalanced Datasets: Fatal accidents (FA) are rare compared to non-fatal accidents (NFA). Standard data mining techniques applied to such imbalanced datasets often produce biased results, overlooking critical factors related to fatalities.
  2. Lack of Spatial Analysis: While many studies identify contributing factors, few integrate spatial analysis to visualize where these accidents are most concentrated, hindering targeted interventions.

This paper aims to bridge these gaps by proposing a robust framework that can handle imbalanced data and provide spatial insights into ROR accident severity.

Methods

The core of the proposed framework consists of four main components:

  1. Data Balancing (BRDB Method): A novel bootstrap-resampling-data-balancing (BRDB) method is introduced to tackle the imbalanced dataset problem. Instead of creating a single balanced dataset, BRDB generates multiple balanced datasets by resampling the majority class (NFA) using bootstrapping. This approach aims to improve the robustness and reliability of the results by reducing the randomness associated with single-dataset balancing.
  2. Association Rule Mining (ARM): The Apriori algorithm is employed to discover relationships between accident characteristics and severity. ARM is chosen for its ability to identify both individual factors (two-item rules) and interactive relationships between multiple factors (three-item rules) without making prior assumptions about data distribution.
  3. Ensemble Method: To further enhance robustness, an ensemble method is used to integrate the rules extracted from the multiple balanced datasets. This involves calculating the mean and confidence intervals of rule parameters (support, confidence, lift) across all generated rules and then selecting the most significant ones based on predefined thresholds.
  4. Geographic Information System (GIS) Analysis: GIS is utilized for spatial analysis to identify "hot spots" of ROR accidents. By mapping the density distribution of accidents and their associated key factors, policymakers can pinpoint high-risk areas requiring immediate attention.

Datasets

The study utilizes a dataset of ROR accidents collected by the Roads Corporation of Victoria, Australia, spanning from 2006 to 2017. After data preprocessing, the dataset comprised 31,940 ROR accidents, with a significant imbalance: 4% Fatal Accidents (FA) and 96% Non-Fatal Accidents (NFA). The dataset includes 22 variables categorized into road characteristics, crash characteristics, human characteristics, and environmental information.

Key Insights and Ideas

  • The Power of Ensemble Balancing: The BRDB method, by creating multiple balanced datasets, significantly improves the reliability of ARM results compared to traditional single-dataset balancing techniques. This is crucial for uncovering meaningful patterns in rare events like fatal accidents.
  • Interactive Factors Amplify Risk: While individual factors are important, three-item rules reveal that the interaction between factors can dramatically increase the likelihood of a fatal outcome. This highlights the need to consider the complex interplay of conditions rather than isolated risks.
  • GIS as a Critical Tool for Targeted Safety: Mapping accident hot spots associated with specific risk factors provides a clear, actionable visual guide for authorities to implement localized and effective safety interventions.

Key Results

  • Six Key Factors for Fatal Accidents: The study identified six individual key factors strongly associated with fatal ROR accidents: 'Helmet/Belt Worn = No', 'Light Condition = Dark street with no lights', 'Types of Off-path Accidents = 8 (Off Left Bend into Object)', 'Speed Limit = 100/110', 'Time of Day = Late in night', and 'Driver Age >= 65'.
  • High Confidence and Lift in Interactions: Three-item rules demonstrated higher confidence and lift levels than two-item rules, indicating that interacting factors significantly increase accident severity. For instance, not wearing a helmet/belt combined with a high-speed limit (100/110) showed a substantially higher probability of fatality.
  • Spatial Hot Spots Identified: GIS analysis revealed that the "Melbourne Metropolitan Area" is a significant hot spot for ROR accidents. Within this area, specific regions were identified as high-risk zones for accidents related to the key FA factors.
  • Robustness Validation: The proposed BRDB method and ensemble approach were shown to provide more robust and reliable results compared to traditional data-balancing methods (under-sampling, over-sampling, mix-sampling), consistently identifying key factors across multiple tests.

Limitations and Future Work

  • Parameter Thresholds: The study acknowledges that the parameter thresholds for ARM (support, confidence, lift) were determined through trial-and-error. Future work could focus on developing more objective methods for setting these parameters.
  • Additional Factors: The authors suggest incorporating more factors in future research, such as traffic flow conditions, population density, and economic conditions, to provide a more comprehensive analysis.
  • Data Fusion: Integrating data from diverse sources with different structures presents a challenge. More advanced data fusion technologies could be explored to enhance the framework's capabilities.

Summary updated

Abstract

Run-off-road (ROR) accidents cause a large proportion of fatalities on roads. Exploring key factors is an effective method to reduce fatalities and improve safety sustainability. However, some limitations exist in current studies: (1) Datasets of ROR accidents have imbalance problems, in which the samples of fatal accidents (FA) are much less than non-fatal accidents (NFA). Data mining methods on such imbalanced datasets make the results biased. (2) Few studies conducted spatial analysis of ROR accidents in visualization. Therefore, this study proposes an association rule mining (ARM)-based framework to analyze ROR accidents on imbalanced datasets. A novel method is proposed to address the imbalance problem and ARM is applied to analyze accident severity. Geographic information system (GIS) is adopted for spatial analysis of ROR accidents. The proposed framework is applied to ROR accidents in Victoria, Australia. Six FA factors and seven NFA factors are identified from two-item rules. The results of three-item rules indicate factors acting interactively increase the likelihood of FA or NFA. Hot spots of ROR accidents are presented by GIS maps. Effective measures are accordingly proposed to improve road safety. Compared with traditional data-balancing methods, the proposed framework has been validated to provide more robust and reliable results on imbalanced datasets.
BibTeX
@inproceedings{jiang2020analysis,
  author = {Feifeng Jiang and Richard K.K. Yuen and Eric Wai Ming Lee and Jun Ma},
  title = {{Analysis of Run-Off-Road Accidents by Association Rule Mining and Geographic Information System Techniques on Imbalanced Datasets}},
  booktitle = {Sustainability},
  year = {2020},
  doi = {10.3390/su12124882},
}
Figure