A machine learning system for classifying real-time urban traffic conditions into four categories — Heavy, High, Normal, and Low — based on vehicle counts, time of day, and day of the week.
Managing urban traffic efficiently requires accurate, real-time insight into road conditions. This project develops a multi-class traffic situation classifier trained on data collected by a computer vision system, enabling automated identification of traffic states to support adaptive signal control and urban traffic management.
Six classification algorithms were evaluated, with Random Forest selected as the final model based on superior accuracy and F1 score. The final model was optimized using Optuna for hyperparameter tuning and trained with SMOTE to address class imbalance, validated through Stratified K-Fold Cross-Validation.
| Class | Label | Description |
|---|---|---|
| 1 | Heavy | Severely congested traffic |
| 2 | High | Above-normal traffic volume |
| 3 | Normal | Typical traffic flow |
| 4 | Low | Minimal traffic |
| Model | Notes |
|---|---|
| Decision Tree | Baseline tree-based classifier |
| Random Forest | Final model — tuned with Optuna |
| Support Vector Machine | Kernel-based classifier |
| K-Nearest Neighbors | Distance-based classifier |
| Gradient Boosting | Ensemble boosting method |
| Naive Bayes | Probabilistic baseline |
