Abstract

This research focuses on comparing different machine learning models to classify handwritten digits in the Devanagari script. The models compared include K-Nearest Neighbors (K-NN), Support Vector Machine (SVM), Convolutional Neural Network (CNN), GoogLeNet (Inception v1), and ResNet-50. GoogLeNet and ResNet-50 are complex deep neural networks typically used for advanced image classification. This study aims to see how well they perform on simpler tasks like Devanagari digit classification. The need for accurate digit classification in India is growing, especially for document scanning, ID card recognition, and digitizing records. The main goal is to find the most accurate and efficient model for this purpose. Surprisingly, the proposed simple CNN model outperforms the more complex GoogLeNet and ResNet-50, achieving an accuracy of 99.522% and an F1 score of 0.9978. Additionally, this CNN model surpasses other CNN models used for Devanagari numeral classification