Comparison of Different Machine Learning Algorithms on the Classification of Handwritten Digits

---

Copy For Citation

Coşar H. İ., Altın C.

2nd International Graduate Studies Congress, Balıkesir, Turkey, 8 - 11 June 2022, pp.75-82

• Publication Type: Conference Paper / Full Text
• City: Balıkesir
• Country: Turkey
• Page Numbers: pp.75-82
• Yozgat Bozok University Affiliated: Yes

Abstract

Recognizing handwritten digits in a computer vision system is a challenging problem

that is critical for a range of new applications. Optical character recognition is a science that

permits the analysis, editing, and search of numerous kinds of texts and pictures. Over the last

few decades, academics have used machine learning methods to automate the analysis of

handwritten and printed documents in order to convert them to electronic format.

Six distinct machine learning methods were used in this work to categorize handwritten digits

from the MNIST dataset: KNN, LDA, QDA, NB, SVM linear, and SVM polynomial. As a

consequence, the QDA algorithm achieved the highest average score of 94.91 percent, whereas

SVM linear had the lowest average score of 85.39 percent.