Driver behavior significantly impacts road safety, serving as a critical factor in traffic crash risks. Human error accounts for a large proportion of crashes, emphasizing the need for targeted interventions. To address this challenge, the i-DREAMS project introduced a “Safety Tolerance Zone (STZ)” framework. This innovative framework is designed to maintain drivers within safe operational boundaries by utilizing both real-time interventions, such as in-vehicle alerts, and post-trip feedback mechanisms, including personalized reports and recommendations. This study introduces and evaluates three hybrid machine learning models—DNN-RF, CNN-LSTM, and RNN-AdaBoost—to classify risky driving behavior into three safety levels: Normal, Dangerous, and Avoidable Accident. The hybrid approach combines the strengths of deep learning and traditional machine learning techniques to enhance predictive accuracy and robustness. To achieve this, a naturalistic driving experiment was conducted in Belgium and the United Kingdom, yielding a comprehensive dataset encompassing 69 drivers, 15,389 trips, and 265,512 minutes of recorded driving data. This dataset reflects diverse driving conditions and behaviors, providing a rich basis for analysis. Among the hybrid models, the Deep Neural Network-Random Forest (DNN-RF) model demonstrated the highest accuracy, achieving approximately 97% in both datasets. Critical driving variables identified as predictors included total travel distance, average speed, harsh acceleration, and harsh braking. To further enhance the interpretability of these machine learning models, the Local Interpretable Model-agnostic Explanations (LIME) algorithm was applied. LIME provided valuable insights into regional differences: harsh acceleration and braking were found to be the most influential factors in predicting risky behaviors in Belgium, whereas trip distance and harsh acceleration were more critical in the UK dataset. These findings underscore the potential of machine learning models to offer actionable insights into the factors contributing to hazardous driving behaviors, allowing authorities and organizations to develop real-time interventions and region-specific strategies.
| ID | pc575 |
| Presentation | |
| Full Text | |
| Tags |
