Handwritten Digit Classification using Neural Networks (MNIST)

📌 Project Overview

This project implements a feedforward neural network (Multilayer Perceptron) using TensorFlow/Keras to classify handwritten digits from the MNIST dataset. The notebook demonstrates a complete deep learning workflow, including data loading, preprocessing, model design, training, and evaluation.

The model learns to classify grayscale images of handwritten digits (0–9) with high accuracy.

---

📂 Dataset

MNIST Handwritten Digits Dataset

• 60,000 training images
• 10,000 test images
• Image resolution: 28 × 28 pixels
• Grayscale images
• Classes: digits 0 to 9

The dataset is loaded directly using Keras:

from tensorflow import keras
(X_train, y_train), (X_test, y_test) = keras.datasets.mnist.load_data()

---

Data Preprocessing

• Images are normalized to the range [0, 1].
• Flattened into 1D vectors to feed into the fully connected layers.
• Dataset is split into training and test sets.

---

Model Architecture

The model is a fully connected feedforward neural network (MLP).

Layer	Type	Units	Activation	Purpose
Input	Flatten	784	–	Converts 2D image to 1D vector
Hidden Layer 1	Dense	128	ReLU	Learns high-level features
Hidden Layer 2	Dense	64	ReLU	Feature refinement
Output Layer	Dense	10	Softmax	Class probability distribution

---

Model Evaluation

After training, the model is evaluated on the unseen test dataset.

✅ Results

Test Accuracy: ~ 97–98%