<![CDATA[Handwritten digit recognition is a process of recognizing and identifying numbers using artificial intelligence algorithms such as Convolutional Neural Network (CNN). The application of handwritten number recognition can be developed and used in identifying postal code numbers on letters, identifying nominal amounts on bank checks and others. However, before carrying out the application, model training is needed on the algorithm that will be used so that number recognition is accurate because the problem faced in recognizing handwritten numbers is that images or written data are diverse and difficult to identify. In this study, the CNN algorithm was used with the SquuezeNet architecture with the MNIST (Modified National Institute of Standards and Technology) dataset which is divided into 60,000 training data and 10,000 test data. The platform used to carry out the training and testing process is Google Colab. Training was carried out 12 times using hyperparameters such as Optimizer namely Adam, SGD, and RMSprop, Learning rate namely 0.1, 0.01, 0.001, 0.0001 and Batch Size 64. Based on the research results from 12 trained models, 1 model was obtained with the best results on the Optimizer namely Adam, Learning rate namely 0.0001 and Batch Size 64 resulting in an accuracy of 99.11%.]]>
Copyrights © 2025
