---
license: mit
tags:
  - image-classification
  - cnn
  - tensorflow
  - keras
  - fashion-mnist
datasets:
  - fashion_mnist
metrics:
  - accuracy
  - precision
  - recall
  - f1
---

# Deep CNN for Fashion MNIST – Stage 3 Model Tuning

This model is part of a structured project focused on building and improving deep convolutional neural networks (CNNs) for clothing item classification using the Fashion MNIST dataset.  
This is **Stage 3**, where architectural tuning with Batch Normalisation and Dropout was introduced to improve performance and generalisation.

## Architecture Summary

- Conv2D layers: 32 → 64 → 128 filters
- Batch Normalisation + ReLU after each
- MaxPooling after each block
- Dropout applied after conv blocks and dense layer
- Dense(128) → Dense(10) with softmax output

## Evaluation Metrics (on Test Set)

| Metric     | Value   |
|------------|---------|
| Accuracy   | 0.9012  |
| Precision  | 0.9053  |
| Recall     | 0.9012  |
| F1 Score   | 0.8992  |

## Dataset

- Fashion MNIST (28x28 grayscale images)
- 60,000 training samples
- 10,000 test samples
- 10 clothing classes

## Author

Alfred Ogunbayo – MSc AI  
GitHub: https://github.com/freddylags  
Hugging Face: https://huggingface.co/alfred-ogunbayo