Evaluation of Classifiers Based on Neural Network for Handwritten Digit Recognition

Handwritten Digit Recognition (HDR) refers to the computer’ s ability to identify and categorize hand written digits from various sources into 10 predetermined categories, ranging from 0 to 9. The issue of digit recognition is significant and crucial. Due to variations in size, thickness, location, and orientation of manually written digits, several challenges need to be addressed in order to evaluate the problem of recognizing handwritten digits. The diversity and variety in the composition techniques of different persons also impact the pattern and appearance of the digits. It refers to the approach of recognizing and organizing written numbers. It offers a diverse array of uses, such as automated bank checks, postal addresses, and tax paperwork, among others. The objective of this research is to employ a classification system to accurately identify handwritten numerals. The consequences of the most widely employed deep learning algorithms, such as multilayered perceptron network, in conjunction with convolution neural networks, are being discussed. Comparison of five neural network classifier shows that accuracy up to 99% can be achieved only in few epochs of training with default configuration of hyper-parameters.