Trafic accidents pose a significant threat to public safety, making their prediction and understanding of contributing factors crucial for efective preventive measures. This study focuses on leveraging historical accident data from Bogotá, Colombia, to design and evaluate machine learning models for trafic accident prediction and take the initial steps toward the identification of the most influential factors associated with each accident. The main objective of this research is to develop accurate machine learning models that can efectively predict trafic accidents in Bogotá and that can later be used in the identification of the key contributing factors leading to these incidents. By achieving this objective, it will be possible to enhance road safety and devise targeted interventions to reduce the occurrence of accidents in the city. A comprehensive dataset comprising historical trafic accident records in Bogotá was collected and preprocessed for analysis. Various machine learning algorithms, including decision trees, random forests, and neural networks, were applied to develop predictive models. The models were trained and evaluated using the F1-score as well as the area under the ROC curve. The experimental results demonstrate the efectiveness of machine learning models in predicting tracfi accidents in Bogotá. The best-performing models achieved performance scores over 0.80, both for the F1 metric and the area under the ROC curve, outperforming traditional statistical methods. These preliminary results are novel in the use of a more comprehensive and updated dataset of accidents in Bogotá and are envisaged to be extended with further analyses in order critical factors that strongly influence accident occurrence.
The increase in road trafic accidents poses a significant challenge to urban areas worldwide,
affecting public safety, transportation eficiency, and overall societal well-being. The development
of efective strategies for accident prevention requires a deep understanding of the
contributing factors as well as an ability to accurately predict accident occurrences [
Bogotá, the capital city of Colombia, is home to a densely populated urban environment
characterized by complex trafic dynamics and diverse transportation modes. Understanding the
factors that contribute to trafic accidents in Bogotá is crucial for designing targeted interventions
and improving road safety initiatives. By harnessing the power of machine learning algorithms,
investigation in this line of research aims to develop accurate prediction models that can assist
policymakers, city planners, and law enforcement agencies in making informed decisions and
allocating resources efectively. Within the local context of Bogotá, previous studies have
made strides toward enhancing the understanding of trafic accident patterns and risk factors.
However, the majority of these investigations relied on historical data that may no longer
accurately reflect the evolving dynamics of the city’s trafic landscape [
The present study builds upon a rich dataset comprising historical accident records collected over the past decade in Bogotá. The dataset encompasses a wide range of variables, both temporal and geographical, and accident severity. By systematically analyzing this comprehensive dataset, we strive to identify the key factors influencing accident occurrences and evaluate the performance of various machine learning models in predicting accidents with high precision and recall rates.
In conclusion, this article contributes to the growing body of research on using machine learning for trafic accident prediction and factor identification. By applying advanced machine learning techniques to historical accident data in Bogotá, we aim to enhance our understanding of the factors influencing accident occurrences and provide valuable insights for policymakers and stakeholders. The results of this study have the potential to inform evidence-based strategies for improving road safety, reducing accident rates, and creating a safer transportation environment in Bogotá and beyond.
Several studies have been conducted worldwide to leverage machine learning techniques for trafic accident prediction and the identification of contributing factors. In this section, we present a review of relevant literature, focusing on similar research eforts and their findings in the context of trafic accident analysis.
[
Following this line of research, [
While these studies above have contributed significantly to the field of trafic accident prediction and factor identification, there is a limited number of studies focusing specifically on Bogotá, Colombia. The present study aims to address this research gap by utilizing a comprehensive dataset of historical accident records from Bogotá and applying a range of machine learning models to predict accident occurrences accurately. Furthermore, our study seeks to identify the most influential factors contributing to trafic accidents in the unique urban context of Bogotá, providing valuable insights for policymakers, transportation authorities, and urban planners in their eforts to improve road safety and reduce accident rates.
Historical trafic accident data for Bogotá, Colombia, spanning the years 2015 to 2022, was sourced from three distinct repositories managed by the local government. These repositories contained detailed information on the vehicles involved, injured individuals, and fatalities associated with each accident. The data, provided in CSV format, was merged into a consolidated dataset based on a unique accident identification number shared across all three repositories. The integrated dataset comprised over 220,000 instances, each encompassing the accident’s temporal attributes (time, year, month, weekday), geographical coordinates (latitude, longitude), accident severity (vehicle damage, injured, death), and the categorical class attribute "Type" (crash, runover, fall from a vehicle). The dimensionality of the integrated dataset was streamlined through the implementation of Principal Component Analysis (PCA). By selecting a subset of factors that collectively explain no less than 95% of the variance, PCA is used to distill the essential features while curtailing redundancy and noise.
Three distinct Machine Learning models were selected for prediction: multilayer neural networks, random forests, and decision trees. These models were chosen due to their capacity to handle complex datasets and demonstrate proficiency in predictive tasks. Prior to training the models, data preprocessing was carried out. This encompassed handling missing values, encoding categorical variables, and normalizing numerical features to ensure consistency and optimal model performance. Model
F1-Score
Area Under ROC Curve Multilayer Neural Network
Random Forest Support Vector Machine
To robustly assess the models’ performance, a 10-fold cross-validation approach was employed. The dataset was divided into ten subsets, with each model trained and evaluated ten times, using a diferent subset as the validation set in each iteration while the rest were used for training. The performance of the prediction models was evaluated using two key metrics: the F1 score and the area under the Receiver Operating Characteristic (ROC) curve. The F1 score ofers a balanced assessment of precision and recall, particularly relevant for imbalanced datasets like this. The ROC curve and its associated area provide insights into the model’s ability to discriminate between classes. Following the cross-validation process, the three models were compared based on their F1 scores and ROC curve areas. This comparison aimed to identify the model that demonstrated the most robust and accurate performance in predicting trafic accidents and classifying their types.
The study strictly adhered to ethical guidelines and data privacy regulations. The utilized data was obtained from publicly available sources and did not contain any personally identifiable information.
All three prediction models exhibited a commendable level of performance, as evidenced by F1-scores and AUC-ROC scores exceeding 0.80. This underscored their eficacy in efectively predicting trafic accidents based on historical data. However, closer scrutiny of the results revealed nuanced distinctions between the models. The multilayer neural network model notably outperformed the others in terms of AUC-ROC score. This outcome signified its superior ability to diferentiate between distinct accident types. Conversely, the support vector machine model demonstrated a slight superiority in terms of F1-score, reflecting its capacity to achieve a harmonious balance between precision and recall.
The tabulated comparison is shown in Table 1 succinctly outlines the contrasting performance of the three models in relation to both the F1-score and the area under the ROC curve. The analysis reafirms the multilayer neural network’s superior AUC-ROC score, while the support vector machine model’s marginally elevated F1-score showcases its prowess in achieving precision-recall equilibrium.
This study addresses a significant gap in the field of trafic accident prediction by leveraging advanced Machine Learning techniques to analyze the most recent and integrated historical accident data provided by the local government of Bogotá. While previous research has explored computational methods for investigating trafic accidents, our work stands out as the first to employ an innovative approach that integrates multiple updated data repositories managed by the city’s government.
The outcomes of this study underscore the potential of machine learning methods in predicting trafic accidents and shedding light on their underlying dynamics. By demonstrating the viability of these approaches, we ofer compelling evidence that such predictive models can provide crucial insights to enhance the local government’s ability to comprehend and address this pressing issue. Our findings not only confirm the feasibility of utilizing machine learning in the realm of trafic safety but also highlight the potential of these models to support the design and implementation of preventive measures aimed at curbing this phenomenon.
It is important to note that while our study presents promising preliminary results, there are avenues for further exploration and refinement. Our current findings serve as a foundational platform for future investigations, where a more comprehensive exploration of the most influential factors contributing to trafic accidents can be undertaken. By integrating additional datasets managed by the local government, such as weather and vehicle conditions, we anticipate that these computational methods will provide even deeper insights into the complex interactions surrounding trafic accidents.
In conclusion, this study marks a significant step forward in the domain of trafic accident prediction and prevention in Bogotá. The pioneering integration of diverse and updated datasets, coupled with the application of machine learning models, has illuminated the potential to tackle this challenge in a novel and efective manner. As we move forward, the insights gleaned from this study will not only inform targeted interventions but also inspire continued research to comprehensively understand and mitigate the causes and consequences of trafic accidents.
The authors would like to express their gratitude to Universidad Antonio Nariño1 for the ifnancial support ofered during the completion of the present investigation. 1Universidad Antonio Nariño (https://www.uan.edu.co/).