Handwritten digit recognition using machine learning

Abstract

Different people are expected to possess varying writing styles that are distinctive to their personalities. There are many aspects that make handwriting differ from person to person, which include spaces, inclination, height, basic patterns, connecting strokes, sizes and widths, markings, and ornaments, among others. However, the challenges come when the handwritten text must be converted into digital form to enhance information sharing and storage. To address this challenge, the recent past has seen the rise of reliance on machines over humans and the subsequent development of machine learning algorithms. Likewise, the recognition of handwritten text is part of the important field of research and development that appear to have a promising future. The handwritten characters recognition has gained considerable attention in the area of machine learning and pattern recognition, most recently. Several techniques have been proposed for recognizing handwritten text. In line with this, many studies have been conducted that explain techniques employed in converting textural substance from paper material to a form that is readable by a machine. Such a digital recognition system has the potential to create a paperless environment by processing existing paper documents and digitization in future. This paper presents two recent techniques based on neural networks, Convolutional Neural Networks (CNNs) and Waikato Environment for Knowledge Analysis (WEKA) and makes a comparative analysis. In this paper, we want to ensure the reliable and effective approaches to the recognition of handwritten digits. The results show that CNN is comparatively more accurate than WEKA, with an accuracy rate as high as 99.59 percent, with the lowest reported degree of accuracy standing at 88.3 percent. On the other hand, the highest report accuracy with regards to WEKA was 82.92 percent, and the lowest stood at 67.45 percent. Although there is a need for more research to be carried out, the information from the two algorithms demonstrates that the existing techniques are increasingly becoming more reliable, and CNN is leading in terms of accuracy.