📌 Overview
This project is focused on classifying handwritten digits using a dense neural network built with TensorFlow and Keras. The model is trained and evaluated on the MNIST dataset, which is
a benchmark dataset for image classification in machine learning. The goal is to classify grayscale images of handwritten digits (0–9) into their
respective categories.
✨ Features
Simple and easy-to-understand neural network
Uses TensorFlow/Keras framework
Fully connected (dense) layers
Input normalization and reshaping
Visualization of sample images and confusion matrix
🧱 Dense Layer Architecture
Input layer: Flattened 28x28 grayscale image to 784-dimensional vector
Hidden layer: Dense layer with 128 neurons and ReLU activation
Output layer: Dense layer with 10 neurons (one per digit) and Softmax activation
📊 Dataset: MNIST Source: https://www.kaggle.com/datasets/hojjatk/mnist-dataset
Training Samples: 60,000
Testing Samples: 10,000
Image Size: 28x28 pixels
Channels: 1 (grayscale)
MNIST consists of 70,000 grayscale images of handwritten digits (0 to 9), each of size 28x28 pixels.
It is divided into 60,000 training images and 10,000 test images.
Each image is labeled with the correct digit it represents.
The pixel values range from 0 to 255 and are normalized to the range 0 to 1 before feeding into the model.
📈 Results
Loss Function: Sparse Categorical Crossentropy
Optimizer: Adam
Epochs: 10
Batch Size: 32
Test Accuracy Achieved: 92%
Confusion Matrix:
📚 References
https://www.kaggle.com/datasets/hojjatk/mnist-dataset
https://www.geeksforgeeks.org/machine-learning/handwritten-digit-recognition-using-neural-network/
https://scikit-learn.org/stable/auto_examples/classification/plot_digits_classification.html
👨💻 Author
Muqadas Ejaz
BS Computer Science (AI Specialization)
Machine Learning & Computer Vision Enthusiast
📫 Connect with me on LinkedIn
🌐 GitHub: github.com/muqadasejaz