Fashion-Class-Classification

👗 Fashion MNIST Image Classification with CNN

---

📌 Table of Contents

• Overview
• Dataset
• Project Structure
• Pipeline
• Model Architecture
• Installation
• Usage
• Results
• Tech Stack
• Known Issues & Fixes
• Future Improvements
• License

---

🧠 Overview

This project implements a Convolutional Neural Network (CNN) from scratch using Keras to classify grayscale images of clothing and fashion accessories from the Fashion MNIST dataset.

The project covers the complete machine learning workflow:

1. Raw data loading from CSV files
2. Data preprocessing and normalization
3. Exploratory data visualization
4. CNN model design and training
5. Evaluation using confusion matrix and classification report
6. Visual prediction display on test images

Fashion MNIST is widely used as a more challenging benchmark than the classic handwritten digits MNIST dataset, making it ideal for practicing image classification with deep learning.

---

🗂️ Dataset

The Fashion MNIST dataset was created by Zalando Research and contains 70,000 grayscale images across 10 fashion categories.

Dataset Statistics

Split	Samples	Image Size	Format
Training	60,000	28 × 28 (grayscale)	CSV
Test	10,000	28 × 28 (grayscale)	CSV

Class Labels

Label	Class Name	Description
0	T-shirt/Top	Short-sleeved or sleeveless top
1	Trouser	Long pants/trousers
2	Pullover	Knitted pullover sweater
3	Dress	One-piece dress garment
4	Coat	Heavy outerwear coat
5	Sandal	Open-toed summer footwear
6	Shirt	Collared button-up shirt
7	Sneaker	Casual athletic shoe
8	Bag	Handbag or tote
9	Ankle Boot	Short boot covering the ankle

📥 Download the dataset from Kaggle. A free Kaggle account is required.

---

📁 Project Structure

fashion-Class-Classification/
│
├── fashion-MNIST-dataset/          # Dataset folder (not tracked by git)
│   ├── fashion-mnist_train.csv     # 60,000 training samples
│   └── fashion-mnist_test.csv      # 10,000 test samples
│
├── visualizing.py                  # Step 1: Data loading & visualization
├── main.py                         # Step 2: Model training & evaluation
│
├── .gitignore                      # Ignores dataset and venv folders
├── requirements.txt                # Python dependencies
└── README.md                       # Project documentation

File Responsibilities

File	Responsibility
visualizing.py	Loads CSVs, converts to NumPy arrays, plots sample grids
main.py	Preprocesses data, builds CNN, trains model, evaluates and visualizes results

---

⚙️ Pipeline

Step 1 — Data Loading (visualizing.py)

fashion_train_df = pd.read_csv('fashion-MNIST-dataset/fashion-mnist_train.csv')
fashion_test_df  = pd.read_csv('fashion-MNIST-dataset/fashion-mnist_test.csv')

training = np.array(fashion_train_df, dtype=np.float32)
test     = np.array(fashion_test_df,  dtype=np.float32)

Each row in the CSV contains:

• Column 0 → class label (integer 0–9)
• Columns 1–784 → pixel values (0–255) flattened from a 28×28 image

---

Step 2 — Visualization (visualizing.py)

• Displays a single random image with its label
• Renders a 15 × 15 grid of random training samples to inspect class distribution and image quality

plt.imshow(training[i, 1:].reshape(28, 28), cmap='gray')

---

Step 3 — Preprocessing (main.py)

# Extract pixels and labels
X_train = training[:, 1:] / 255.0    # Normalize pixels to [0, 1]
y_train = training[:, 0]             # Integer class labels (0–9), no normalization

# 80/20 train-validation split
X_train, X_valid, y_train, y_valid = train_test_split(
    X_train, y_train, test_size=0.2, random_state=12345
)

# Reshape for CNN input: (samples, height, width, channels)
X_train = X_train.reshape(X_train.shape[0], 28, 28, 1).astype('float32')
X_valid = X_valid.reshape(X_valid.shape[0], 28, 28, 1).astype('float32')
X_test  = X_test.reshape(X_test.shape[0],  28, 28, 1).astype('float32')

Step	Detail
Normalization	Divide pixels by 255 → range [0, 1]
Validation split	80% training, 20% validation
Reshape	(n, 784) → (n, 28, 28, 1) for CNN

---

Step 4 — Model Training (main.py)

model.compile(
    optimizer=Adam(learning_rate=0.001),
    loss='sparse_categorical_crossentropy',
    metrics=['accuracy']
)

model.fit(
    X_train, y_train,
    validation_data=(X_valid, y_valid),
    epochs=10,
    batch_size=32,
    verbose=1
)

---

Step 5 — Evaluation (main.py)

# Get predicted class indices from softmax probabilities
predictions = np.argmax(model.predict(X_test), axis=1)

# Confusion matrix heatmap
cm = confusion_matrix(y_test, predictions)
sns.heatmap(cm, annot=True, fmt="d")

# Per-class precision, recall, F1-score
print(classification_report(y_test, predictions, target_names=target_names))

---

🏗️ Model Architecture

Input: (28, 28, 1)
       ↓
Conv2D         → 32 filters, 3×3 kernel, stride=2, ReLU
       ↓
MaxPooling2D   → 2×2 pool size
       ↓
Flatten        → 1D vector
       ↓
Dense(32)      → ReLU activation
       ↓
Dense(10)      → Softmax (output: probability per class)

Layer Summary

Layer	Output Shape	Parameters	Notes
Conv2D	(13, 13, 32)	320	3×3 kernel, stride 2, ReLU
MaxPooling2D	(6, 6, 32)	0	2×2 pool
Flatten	(1152,)	0	—
Dense	(32,)	36,896	ReLU activation
Dense	(10,)	330	Softmax output

Hyperparameters

Hyperparameter	Value
Optimizer	Adam
Learning rate	0.001
Loss function	Sparse Categorical Crossentropy
Epochs	10
Batch size	32
Validation split	20%

---

🛠️ Installation

Prerequisites

• Python 3.9 or higher
• pip

1. Clone the repository

git clone https://github.com/your-username/fashion-Class-Classification.git
cd fashion-Class-Classification

2. Create a virtual environment (recommended)

python -m venv .venv

# Activate on Windows
.venv\Scripts\activate

# Activate on macOS/Linux
source .venv/bin/activate

3. Install dependencies

pip install -r requirements.txt

Or manually:

pip install pandas numpy matplotlib seaborn keras tensorflow scikit-learn

4. Download the dataset

1. Visit https://www.kaggle.com/datasets/zalando-research/fashionmnist
2. Click Download (free Kaggle account required)
3. Extract and place the CSV files as shown:

fashion-MNIST-dataset/
├── fashion-mnist_train.csv
└── fashion-mnist_test.csv

---

▶️ Usage

Run visualization only

python visualizing.py

Outputs:

• One random training image with its class label
• A 15×15 grid of random training images with labels

Train and evaluate the model

python main.py

Outputs:

• Live training progress (accuracy & loss per epoch)
• 5×5 prediction grid: predicted vs true labels
• Confusion matrix heatmap
• Full classification report in terminal

---

📊 Results

After 10 training epochs, the model produces the following evaluations:

Confusion Matrix

A 10×10 annotated heatmap where rows are true labels and columns are predicted labels. A strong diagonal means high accuracy per class. Off-diagonal values reveal which classes are most commonly confused.

Classification Report

Per-class breakdown of key metrics:

Metric	Description
Precision	Of all predictions for a class, how many were correct
Recall	Of all true instances of a class, how many were found
F1-Score	Harmonic mean of precision and recall
Support	Number of true instances per class in the test set

Prediction Grid

A 5×5 visual grid displaying test images with their predicted and true labels side by side for quick qualitative review.

---

📦 Tech Stack

Library	Purpose
pandas	Loading CSV dataset files
numpy	Array operations and image reshaping
matplotlib	Plotting images and sample grids
seaborn	Confusion matrix heatmap visualization
keras	Building, compiling, and training the CNN
tensorflow	Keras backend for computation
scikit-learn	Train/test split, confusion matrix, report

---

🐛 Known Issues & Fixes

Issue	Root Cause	Fix Applied
FileNotFoundError on CSV load	Wrong path or missing dataset files	Verified fashion-MNIST-dataset/ folder structure
White/blank image grid	Missing cmap='gray' in imshow	Added cmap='gray' to all imshow calls
ValueError: cannot reshape array of size 1	training[i, 1] grabs 1 pixel instead of 784	Changed to training[i, 1:] (slice, not index)
TypeError: tuple cannot be interpreted as int	X.shape[0:] returns full tuple, not int	Changed to X.shape[0] (no colon)
confusion_matrix receives None	prediction_model() had no return statement	Added return predict_classes at end of function
Labels divided by 255 incorrectly	y_train = training[:, 0] / 255	Labels are integers 0–9 and must NOT be normalized
Input layer placed after Conv2D	Misplaced model.add(Input(...)) mid-model	Removed stray Input layer; Keras infers shape from input_shape
Wrong loss function	binary_crossentropy used for 10-class task	Changed to sparse_categorical_crossentropy
Wrong output layer	Dense(16, activation='sigmoid')	Changed to Dense(10, activation='softmax')

---

🔮 Future Improvements

• Add additional Conv2D + MaxPooling2D blocks for deeper feature extraction
• Add BatchNormalization layers to stabilize and speed up training
• Add Dropout layers to reduce overfitting
• Implement learning rate scheduling (e.g. ReduceLROnPlateau)
• Plot training history — accuracy and loss curves per epoch
• Save and reload trained model using .keras or .h5 format
• Build an interactive demo with Gradio or Streamlit for live predictions
• Apply data augmentation (random flips, rotations, zoom) to improve generalization
• Benchmark against other architectures (ResNet, MobileNet via transfer learning)

---

📄 License

This project is licensed under the MIT License.
Feel free to use, modify, and distribute with attribution.

---

Made with 🧠 and ☕  |  Fashion MNIST Classification Project