MNIST Handwritten Digits Classification Using CNN

MNIST Handwritten Digits Classification Using CNN

Assistant Professor Mr. Balaji C*1, Mr. Chandhiran S*2

*1Assistant Professor, Department of Commerce with Business Analytics, Dr. N.G.P. Arts and Science College, Coimbatore, Tamil Nadu, India.

*2Student, Department of Commerce with Business Analytics, Dr. N.G.P. Arts and Science College, Coimbatore, Tamil Nadu, India.

Abstract

A machine learning model was developed for handwritten digit recognition using Convolutional Neural Networks (CNNs). CNNs effectively classify images by learning patterns, outperforming traditional neural networks. A dataset of 1,000 custom images was created, with 80% for training and 20% for testing. Images were preprocessed (cleaning, scaling, reshaping) to optimize performance. A CNN model, built with TensorFlow and Keras, achieved 92.5% accuracy in recognizing digits (0–9). Key libraries include NumPy, Pandas, TensorFlow, and Keras. The study highlights CNNs' advantages in accuracy and efficiency, making them ideal for OCR, banking, and automated document processing.

Keywords: Machine Learning, Handwritten Digit Recognition, Convolutional Neural Networks (CNNs), Image Classification, Optical Character Recognition (OCR), Automated Document Processing.