Digit Recognition

A neural network-based digit recognition system using the MNIST dataset to classify handwritten digits with preprocessing and visualization

Comment

This project is a neural network-based digit recognition system that utilizes the MNIST dataset to classify handwritten digits (0–9). The dataset comprises 60,000 training and 10,000 testing images of 28x28 grayscale pixels, each representing a single digit. The project includes data preprocessing, model design, training, evaluation, and visualization, providing a comprehensive pipeline for solving this classification problem.

Key Metrics:

  • Accuracy: 97.50%
  • Loss: ~0.14

Features:

  • Classifies digits from 0 to 9 using the MNIST dataset
  • Data Preprocessing:
  • Flattens 3D image data into 2D arrays
  • Normalizes pixel values to the range [-1, 1]
  • Neural Network Model
  • Model Training
  • Trains the neural network using the Adam optimizer
  • Utilizes Sparse Categorical Crossentropy loss function
  • Calculates model loss and accuracy on the test dataset
  • Prediction Visualization

Technologies:

  • Python
  • TensorFlow/Keras
  • MNIST Dataset
  • Matplotlib
  • NumPy

Built With

Share this project:

Updates