Study of Classification Algorithms for Handwritten Character Recognition

Recognizing handwritten characters automatically has a lot of applications and is not an easy task given the number of ways a character is written by people across the world. To automate, this process involves intelligent machine learning algorithms. In this work, the main focus is on using convolutional neural networks (CNN) for efficient character identification. The other traditional algorithms like Naive Bayes and decision tree were also fitted after proper scaling. The CNN model makes use of convolutional, max-pooling, dense, and dropout layers. ReLU was used as activation for all layers except for the last dense layer which uses the softmax activation function. CNN has an advantage over its prior models that is to detect the important features without human supervision. The user input image for the English alphabets was preprocessed by grayscaling, reshaping, sharpening, and resizing. The model takes preprocessed input and gives its prediction for the characters. Experimental results proved that the CNN model provided an accuracy of 93% compared with the other models.