Addressing Data Imbalance in Freeway Traffic State Classification: An FCM-RF-SMOTE Framework with Vehicle Trajectory Data

This study proposes the FCM-RF-SMOTE framework to resolve data imbalance in real-time freeway traffic state classification, integrating Fuzzy C-Means (FCM), Random Forest (RF), and Synthetic Minority Over-sampling (SMOTE). Traffic states are classified by quantitative thresholds: smooth (>110 km/h, <30 veh/km), stable (80–110 km/h, 30–60 veh/km), congested (40–80 km/h, 60–100 veh/km), and severely congested (<40 km/h, >100 veh/km). The framework increases severe congestion representation from 3.67% to 19.83%, enhancing classification accuracy from 77.67% to 97.80%. Validation employs SUMO simulation with Gaussian noise (σ = 0.1 m) and 10 Hz sampling to approximate radar characteristics, with future physical sensor validation required. Results demonstrate robustness for traffic monitoring systems.