ANN

🧠 MNIST Digit Classification with Artificial Neural Network (ANN)

This project builds an Artificial Neural Network (ANN) using TensorFlow/Keras to classify handwritten digits from the MNIST dataset. An ANN is the simplest form of deep learning — layers of interconnected neurons that learn patterns from data.

---

📁 Dataset Overview

The MNIST dataset contains 70,000 grayscale images of handwritten digits (0-9):

• Training set: 60,000 images
• Test set: 10,000 images
• Each image is 28x28 pixels (784 features when flattened)
• 10 classes: digits 0 through 9

---

🎯 Objective

To build a simple feedforward neural network that learns to recognize handwritten digits from pixel values alone, demonstrating the core concepts of deep learning — layers, activation functions, and backpropagation.

---

🧠 Why ANN?

• ANNs are the foundation of deep learning — understanding them is essential before CNNs or RNNs
• A simple 2-3 layer network can achieve >97% accuracy on MNIST
• ANN introduces key concepts: forward propagation, loss functions, optimization, and epochs
• Starting simple makes the jump to CNNs more understandable

---

🧪 Tools & Libraries Used

• Python
• Jupyter Notebook
• pandas, numpy for data handling
• matplotlib, seaborn for visualization
• tensorflow / keras for building and training the neural network

---

📊 Model Architecture

• Input Layer: 784 neurons (28x28 flattened)
• Hidden Layer 1: 128 neurons, ReLU activation
• Dropout: 20% for regularization
• Hidden Layer 2: 64 neurons, ReLU activation
• Output Layer: 10 neurons, Softmax activation

Total parameters: ~110,000

---

📊 Process Overview

1. Data Loading & Exploration:

   • Loaded MNIST from tf.keras.datasets
   • Visualized sample digits
   • Checked class distribution

2. Data Preprocessing:

   • Normalized pixel values to [0, 1] range
   • One-hot encoded labels for multi-class classification
   • Flattened 28x28 images to 784-length vectors

3. Modeling:

   • Built a Sequential model with Dense layers
   • Used categorical crossentropy loss and Adam optimizer
   • Trained for 10 epochs with validation split

4. Evaluation & Visualization:

   • Evaluated on test set
   • Plotted training/validation accuracy and loss curves
   • Visualized sample predictions

---

📌 Key Results

• The ANN achieved >97% accuracy on test data after 10 epochs
• Training and validation accuracy converged smoothly — minimal overfitting
• Most misclassifications occurred between visually similar digits (4 vs 9, 3 vs 8)
• Dropout regularization helped prevent overfitting

---

📂 Folder Structure

ML-DL-Projects/
├── Deep-Learning/
│   └── ANN/
│       ├── dataset/           # (MNIST loaded via tf.keras, no CSV)
│       ├── images/
│       │   ├── training_history.png
│       │   └── sample_predictions.png
│       ├── notebooks/
│       │   └── ANN_Example.ipynb
│       └── README.md
├── uv.lock
├── pyproject.toml

---

🚀 Getting Started

To run this project locally:

Step 1: Set up environment

cd ML-DL-Projects
uv sync

Step 2: Run notebook

jupyter notebook

Then open ANN_Example.ipynb.

---

💡 Future Ideas

• Add more hidden layers or increase neurons and see how accuracy changes
• Experiment with different activation functions (tanh, leaky ReLU)
• Try different optimizers (SGD, RMSprop) and compare convergence
• Compare ANN performance directly with CNN on the same dataset

---

📬 Contact

Feel free to reach out if you have questions or want to collaborate!