Machine Learning Techniques for Handwritten Digit Recognition

Abstract

The following study examines how various classification algorithms perform on the problem of handwritten digit recognition. The classifiers discussed are k-Nearest Neighbour (k-NN), Single Classification Decision Trees and Bagged Decision Trees. These algorithms were evaluated with the use of information from the United States Postal Service (USPS). This study’s results show that the k-NN classifier had the fastest performance while the bagged decision trees were the slowest. In terms of classification performance, the bagged decision tree method was found to have the fewest misclassifications and outperformed k-NN and single classification trees in all of the considered metrics.