Skills and Insights Gained

In this project, my team and I explored various machine learning approaches for traffic sign classification for the course TKPR221 Fundamentals of Machine Learning. We utilized the Mapillary Traffic Sign Dataset and implemented different models to find the best-performing one. Our key steps included:

- Preprocessing the dataset by cropping and resizing images,
- Handling class imbalance with SMOTE,
- Implementing classical ML models like SGD Classifier,
- Building CNN architectures for deep learning-based classification,
- Using ensemble learning techniques to improve results, and
- Fine-tuning ResNet50 for transfer learning to achieve optimal accuracy.

The results demonstrated that transfer learning provided the most consistent and highest accuracy. The project helped us gain hands-on experience with real-world datasets and machine learning model evaluation. You can explore the full implementation and results in our GitHub repository linked above.