Deep Convolutional Neural Networks Transfer Learning Comparison on Arabic Handwriting Recognition System

Around 27 languages and more than 420 million people worldwide use Arabic letters. That makes the language one of most used languages. However, has a challenge, namely difference in letters based on their position. handwriting recognition is important for various applications, such as education communication. One example during pandemic when turned digital, making recognizing students' difficult. This paper aims to create model that can recognize by comparing several CNN architectures using transfer learning classify Arabic, Hijja, AHCD datasets. Transfer been trained previous datasets other suitable models with small because it improve accuracy even The were split into 60%, 20%, 20% training, validation, testing. Each uses data augmentation 50% dropout fully connected layer reduce overfitting. Some this study writing are ResNet, DenseNet, VGG16, VGG19, InceptionV3, MobileNet. compiled parameters. best achieved Hijja dataset VGG16 Adam optimizer 0.0001 rate. Based research, expected know performance classifying handwriting.