Abstract:
Road traffic accidents continue to pose a global public health challenge, resulting in over 1.3
million deaths annually and causing significant economic damage. Current road safety
measures and mobile solutions are often reactive and fail to leverage real-time environmental
and behavioural data. This project addresses the critical need for a proactive, mobile-based
accident prevention system that can anticipate and alert drivers to high-risk situations through
the integration of multiple sensor inputs and predictive analytics.
The project employed a design science research methodology, combining literature review,
system design, machine learning implementation, and prototype development. The mobile
application was developed using Flutter and Firebase, integrating GPS, accelerometer, and
weather APIs. A predictive model using Random Forest was trained on historical and simulated
driving behaviour data to classify risk levels and trigger real-time alerts. Sensor data fusion was
applied to ensure robust feature extraction and context awareness.
Initial results demonstrate that the prototype achieved promising accuracy in real-time risk
detection, with a confusion matrix showing 87% precision, 82% recall, and an F1-score of
84%. The AUC-ROC score was 0.91, indicating high classifier performance. Usability testing
with 10 participants also revealed positive reception, especially regarding real-time alerts and
blackspot detection. Further improvements are planned for interpretability, iOS support, and
cloud-based scalability.
