🧠 MNIST CNN Classifier with Data Augmentation

A robust Convolutional Neural Network (CNN) model trained on the classic MNIST dataset with data augmentation for enhanced generalization. Built with TensorFlow and Keras.

---

🚀 Features

• 🔢 Classifies handwritten digits (0-9) using a CNN
• ↻ Enhanced with Image Data Augmentation (rotation, shifting, zooming)
• 📊 Evaluated with precision, recall, F1-score, and confusion matrix
• 📂 Easily save and reload your trained model
• 🧪 Includes test set evaluation and predictions

---

🧰 Tech Stack

• Python 🐍
• TensorFlow / Keras
• NumPy
• scikit-learn (for evaluation metrics)
• Matplotlib (optional: for visualizations)

---

🖼️ Data Augmentation Techniques

• ↻ Rotation: ±15°
• ↔️ Horizontal/Vertical Shift: 10%
• 🔍 Zoom: ±10%
• ❌ No Flipping (digits are orientation-sensitive)

---

🏗️ Model Architecture

Input (28x28x1)
│
├── Conv2D (32 filters, 3x3, ReLU)
├── MaxPooling2D (2x2)
├── Conv2D (64 filters, 3x3, ReLU)
├── MaxPooling2D (2x2)
├── Flatten
├── Dense (128, ReLU)
└── Dense (10, Softmax)

---

📊 Performance Metrics

After training for 10 epochs:

• ✅ Accuracy: ~99%
• 📋 Precision, Recall, F1-score: Available via classification_report()
• 🔍 Confusion Matrix: Shows per-digit prediction accuracy

---

📦 Getting Started

git clone https://github.com/yourusername/mnist-cnn-augmented.git
cd mnist-cnn-augmented
pip install -r requirements.txt

Run the model training:

python mnist_cnn_augmented.py

---

📊 Evaluation

To evaluate and generate a classification report:

from sklearn.metrics import classification_report, confusion_matrix

y_pred = model.predict(x_test)
y_pred = np.argmax(y_pred, axis=1)

print(classification_report(y_test, y_pred))
print(confusion_matrix(y_test, y_pred))

---

📂 Saving and Loading the Model

# Save
model.save("mnist_cnn_augmented_model.h5")

# Load
from tensorflow.keras.models import load_model
model = load_model("mnist_cnn_augmented_model.h5")

---

👑 Author

S.Arun Prakash
📧 arunprak3@gmail.com

---

🧠 Fun Fact

The MNIST dataset was used to test early neural networks — and it's still a favorite for testing deep learning prototypes today!