MNIST Handwritten Digit Classification (Dense Neural Network)

Project Overview

This project classifies handwritten digits (0–9) using a fully connected neural network (MLP).
The MNIST dataset contains 70,000 grayscale images of handwritten digits.

Key Concepts

• Multi-class Classification
• Fully Connected Neural Network
• Image Preprocessing & Normalization
• Softmax Output Layer
• Sparse Categorical Crossentropy
• Training Curve Visualization

Technologies Used

• Python
• TensorFlow / Keras
• NumPy
• Pandas
• Matplotlib

Files in This Repository

• MNIST(Dense NN).ipynb — Model building, training, evaluation
• requirements.txt — Dependencies
• /assets — Optional folder for accuracy/loss plots

Model Architecture

• Flatten
• Dense(128) → ReLU
• Dense(64) → ReLU
• Dense(10) → Softmax

Loss: Sparse Categorical Crossentropy
Optimizer: Adam

Results

• Test Accuracy: ~97.24%
• Visualized training vs validation curves
• Displayed sample predictions using softmax probabilities

How to Run

MNIST-Digit-Classifier-NN