Sekar Sakti, Lintang Maulida Sekar Bawono, Fitri Trapsilawati

Development of an Artificial Neural Network Model for Predicting Speeding Behaviour: A Case Study from Indonesia

Introduction

Development of an artificial neural network model for predicting speeding behaviour: a case study from indonesia. Develops an Artificial Neural Network (ANN) model to predict speeding behavior on Indonesian intercity roads, identifying high-risk drivers and improving road safety with 86.67% accuracy.

Abstract

Traffic accidents are the third leading cause of death in Indonesia, with speeding behavior being the predominant human factor responsible for most fatal outcomes. Early detection of drivers’ propensity to speed is therefore essential for effective prevention strategies. This study develops an Artificial Neural Network (ANN) model to predict the tendency of drivers to speed on intercity roads using a labeled questionnaire dataset comprising 14 input variables. The dataset was divided into training, validation, and testing subsets, where the validation set was used for hyperparameter tuning, while the testing set was used for final evaluation on unseen data. The model was trained using the Adam optimizer with a binary cross-entropy loss function. The optimal configuration consists of a single hidden layer with 12 neurons using ReLU activation, a Sigmoid output layer, 750 training epochs, and a learning rate of 0.03. The final model achieved an accuracy of 86.67% and a Cohen’s kappa value of 0.7339, which indicates strong predictive reliability. These findings demonstrate the model’s potential as a valuable tool for relevant stakeholders to identify high-risk drivers and design targeted interventions. As a result, the model can be used to proactively reduce speeding-related traffic accidents and improve road safety on intercity routes.

Review

This study tackles a critical public safety issue in Indonesia, where speeding is a predominant human factor in fatal traffic accidents. The authors present an Artificial Neural Network (ANN) model designed to predict drivers' propensity to speed on intercity roads, aiming for early detection and prevention. Leveraging a labeled questionnaire dataset with 14 input variables, the model was trained, validated, and tested, demonstrating robust performance. With an achieved accuracy of 86.67% and a strong Cohen’s kappa value of 0.7339, the findings suggest the model's significant potential as a valuable tool for stakeholders to identify high-risk drivers and consequently improve road safety. The research exhibits several notable strengths, particularly its focus on a pressing real-world problem with clear societal impact in Indonesia. The methodological approach using an ANN is well-suited for predictive modeling, and the abstract provides sufficient detail regarding the model's configuration, including the use of Adam optimizer, binary cross-entropy loss, a single hidden layer with 12 ReLU neurons, Sigmoid output, 750 epochs, and a learning rate of 0.03. The rigorous splitting of the dataset into training, validation, and testing subsets, coupled with the impressive performance metrics, underscores the model's strong predictive reliability and its potential for practical application in targeted intervention strategies. Despite its promising results, the abstract leaves some key aspects requiring further elaboration for a comprehensive review. Crucially, a description of the 14 input variables from the questionnaire is absent; understanding the nature of these features (e.g., demographic, psychological, behavioral) is vital for assessing the model's interpretability and real-world applicability. Additionally, clarification on how "speeding behavior" was defined and labeled in the questionnaire would strengthen the validity of the ground truth. Future research could benefit from exploring feature importance to provide actionable insights for intervention design, comparing the ANN's performance against other machine learning or statistical models, and discussing the practical implementation challenges and ethical considerations involved in deploying such a predictive system by relevant stakeholders.

Full Text

Comments

You need to be logged in to post a comment.