After the COVID-19 epidemic, road traffic accidents remain a major cause of death worldwide. In the post-COVID era, compliance with traffic laws has declined, despite progressive containment of the pandemic through vaccination and other mitigation strategies. Weak enforcement of traffic laws has contributed to increased behaviours such as speeding, driving under the influence and failure to wear seat belts, resulting in fatal road accidents. Most existing studies rely on image- or video-based accident detection techniques, which are expensive and computationally intensive. This study proposes an Extreme Gradient Boosting (XGBoost) model trained on mobile device sensor data as a cost-effective accident detection approach. The model uses accelerometer, gravitational and audio features that can be readily obtained from common mobile devices. A total of 1,700 sensor-based observations were simulated using an Android device from approximately 300 motion and sound scenarios, including rotation, shaking, bouncing and sound levels between 20 and 110 decibels. The XGBoost model achieved 100% accuracy, precision and recall on the simulated data and 99.41% accuracy on an external in-car monitoring dataset from the Data World repository, outperforming existing methods. These results demonstrate the suitability of the proposed model for real-time accident detection using mobile-generated sensor data.
Alu et al. (Sun,) studied this question.