03_mnist_digit_recognition

🔢 Digit Recognition Using MNIST (ANN - MLP)

This project demonstrates how to classify handwritten digits (0 to 9) from the MNIST dataset using a multi-layer perceptron (MLP), a basic form of an Artificial Neural Network (ANN).

---

🧠 Problem

Given an image of a handwritten digit (28x28 grayscale), predict the correct digit label (0–9).

---

🧩 Concepts Used

• Multi-Layer Perceptron (MLP)
• Softmax Output Layer
• Cross-Entropy Loss
• ReLU Activation
• Data Normalization
• One-Hot Encoding
• Model Evaluation

---

🏗️ Model Architecture

Layer	Type	Units	Activation
Input	Flatten	784	-
Hidden 1	Dense	128	ReLU
Hidden 2	Dense	64	ReLU
Output	Dense	10	Softmax

---

🛠️ Files

• train.ipynb: Model building, training, evaluation, and accuracy plots
• model.ipynb: Modular version with prediction and visualization
• mlp_weights.h5: (optional) for loading pretrained weights

---

📈 Results

• Training Accuracy: ~98%
• Test Accuracy: ~97%
• Fast convergence with only 10 epochs

---

📊 Sample Output