In this project, I developed an image classification system capable of distinguishing between five different classes. Our dataset comprises 2400 images, divided into training, validation, and test sets with counts of 1720, 430, and 250, respectively. A key challenge addressed in this project is the imbalance in the dataset, for which we employed techniques like undersampling and oversampling.
- Logistic Regression
- Neural Network
- AlexNet
- ResNet-34
- Inception Net
- Custom CNN Model
Each model was evaluated on both balanced and unbalanced versions of the dataset. We also utilized several augmentation techniques, including rotation, width and height shifts, shear transformations, and horizontal flips to enhance our training data.
Our evaluation focused on metrics such as precision, recall, F1-score, and overall accuracy to assess the performance of each model comprehensively.
This project highlights the effectiveness of different machine learning models in image classification tasks and the impact of dataset balancing techniques on model performance. We hope our findings will contribute to the broader understanding of machine learning applications in image classification.
- Python
