SafeRoute AI - Real-Time Road Crash Risk Prediction
Research pitch for a CNN-XGBoost ensemble that forecasts road crash risk across Melbourne Metro in real time.
Overview
SafeRoute AI is a research pitch for Deakin's Research and Development in IT unit, proposing a real-time crash-risk prediction system for Melbourne Metro that fuses live VicRoads, Bureau of Meteorology and CrashStats data through a CNN-XGBoost ensemble model.
The problem
Current intelligent transport systems are mostly reactive, responding to crashes rather than forecasting them, and most crash-prediction models train on static offline datasets disconnected from live traffic, weather and road-condition feeds - leaving a gap in real-time, location-specific crash-risk forecasting. The WHO reports 1.19 million road deaths worldwide every year, and Victoria alone records hundreds of fatalities annually.
My approach
Proposed a CNN-XGBoost ensemble pipeline ingesting live VicRoads, Bureau of Meteorology and CrashStats Victoria feeds to output multi-class crash-severity predictions (minor/serious/fatal) at 0.8s latency, and designed a live risk-heatmap dashboard concept for Melbourne Metro targeting 92%+ model accuracy - grounding the design in peer-reviewed research on SMOTE class balancing, CNN spatial feature extraction and XGBoost severity classification.