Handwriting Detection using Neural Network

Abstract

This undertaking report compares three models for handwriting detection using MNIST and EMNIST dataset. The result is a system which can recognize any handwritten number or alphabet. The 3 models which we compared are artificial neural network, random forest and XGBoost. We additionally experimented with different hyperparameters to maximize test accuracy and reduce overfitting as much as we could. MNIST and EMNIST are most common dataset available on the internet for handwriting detection. MNIST dataset contains 60000 training images and 10000 test images. EMNIST dataset contains 124800 training images and 20800 test images.