Fashion MNIST Overfitting Analysis

This project demonstrates a deep learning pipeline on the Fashion MNIST dataset to explore overfitting and its mitigation strategies. It presents experimental insights into when and why overfitting occurs and compares the performance of various training techniques in improving model generalization.

📁 Project Structure

• fashion_mnist_overfitting_prevention.ipynb: Main notebook containing model training, evaluation, and overfitting analysis.
• conclusion.txt: Summary of findings and insights.
• .gitignore: Specifies files and directories to be ignored by Git.
• requirements.txt: Lists all Python dependencies.

🔍 Key Questions Answered

1. Can you create and document a scenario where overtraining occurred?
   Yes, training for 50 epochs without regularization led to training accuracy reaching 99% while validation accuracy plateaued, indicating overfitting.

2. What training methods did you find helpful to prevent overtraining and why?

   • Data augmentation
   • Early stopping
   • Learning rate scheduling

3. Did you reach a point where it was clear you should stop training?
   Yes, training was stopped at epoch 14 based on validation loss behavior.

4. What image classes had the best and worst performance?

   • Best: Trouser (Class 1), Bag (Class 8)
   • Worst: Shirt (Class 6)

   Improvement Recommendation: Apply class-specific augmentation to emphasize distinguishing features.

🚀 Getting Started

1. Clone the repository

git clone https://github.com/your-username/fashion-mnist-overfitting.git
cd fashion-mnist-overfitting

2. Set up the environment

pip install -r requirements.txt

3. Run the notebook

Open fashion_mnist_overfitting_prevention.ipynb using Jupyter or VSCode and run the cells step by step.

📦 Requirements

See requirements.txt for full list.

📜 License

This project is licensed under the MIT License.