A Case Study in Digit Recognition

Abstract

Digit Recognition is a computer vision technique to predict the numerical value of digits in a dataset. The issue is that handwritten digits are not always the same size and that these digits are not always written the same way from person to person. The goal of this project is to take an image of a handwritten image single digit and determine what it is. The objective is to build an efficient model with an accuracy of at least 90% through testing and training on a Kaggle dataset. We will explore various machine learning algorithms such as SVM and K nearest neighbors as to whether they improve the efficiency of recognizing the hand-drawn digits. We present our findings through various metrics and graphical representations.