HANDWRITTEN DIGIT RECOGNITION USING MACHINE LEARNING AND DEEP LEARNING TECHNIQUES: A COMPARATIVE STUDY OF SVM, KNN, RFC, AND CNN MODELS

Abstract

Handwritten digit recognition is a significant and challenging problem in the field of pattern recognition and machine learning. The variability in digit size, thickness, orientation, and positioning—combined with the diverse writing styles of individuals—introduces complexity in accurately identifying digits. This task plays a crucial role in numerous real-world applications such as automated check processing, postal address interpretation, and tax form digitization.This project focuses on implementing classification algorithms to recognize handwritten digits. It explores the performance of several well-known machine learning models including Support Vector Machines (SVM), K-Nearest Neighbors (KNN), and Random Forest Classifiers (RFC), as well as a deep learning approach using a multilayer Convolutional Neural Network (CNN) built with Keras and powered by Theano and TensorFlow backends. Experimental results demonstrate that the CNN model achieved the highest accuracy of 98.70%, outperforming traditional algorithms such as SVM (97.91%), KNN (96.67%), and RFC (96.89%).