An Intelligent Handwritten Digits and Characters Recognition System

The process of giving machines the ability to recognize human handwritten digits and characters is known as handwritten digit and character recognition. Handwritten digits and characters are imperfect, vary from person to person, and can be constructed with a variety of flavors. Therefore, it's not a simple assignment for the machine. In this paper, a machine learning algorithm has been made to detect handwritten digits and characters with high accuracy relative to the past models. The MNIST dataset is used to provide the model with the training and test datasets for its variety of data. Another dataset is Kaggle's A-Z Handwritten Alphabets dataset, which is used for the English letter’s dataset. It contains a total of 370,000 images of English letters. The paper's output is a Graphical user interface (GUI) that allows users to draw a digit or character and instantly see a digital version of it along with an accuracy %. The convolutional neural network (CNN) approach was applied as a deep learning method. The suggested CNN model is based on the Keras model, which classifies handwritten digit pictures using an RMSprop optimizer. With epoch 10, the suggested CNN model achieves 98.80 percent accuracy during testing and 99.06 percent accuracy during training. Average macro accuracy of 0.99 was attained. It has a 0.98 average macro recall. The macro average F1 score reached 0.99.