<![CDATA[This research aims to develop an automatic classification model to recognize the type of Arabic calligraphy writing using MobileNetV2 Convolutional Neural Network (CNN) architecture. Arabic calligraphy has a visual uniqueness and complexity of letterforms that become a challenge in the classification process, especially for ordinary people. The four main calligraphy types used in this research are Tsulust, Naskhi, Diwani, and Kufi. The research follows the CRISP-DM stages which include business understanding, data understanding, data preparation, modeling, evaluation, and deployment. The dataset used is the HICMA dataset consisting of 5,031 calligraphy images. The data is processed through cropping, normalization, and resizing to 224x224 pixels. The model was trained with epoch variations (10, 20, 30, and 40) to obtain the best configuration. The results show that the model at the 20th epoch has the most optimal performance with a testing accuracy of 97.52%. Evaluation of classification metrics showed high F1-Score values in the majority classes. The previously low-performing Kufi class was improved through data augmentation techniques to obtain an F1-Score value of 0.99. The model is then integrated into a Flask-based web application that allows users to upload images and receive classification results directly. The results of this research show that MobileNetV2 is effective for Arabic calligraphy type classification and can be practically implemented for educational purposes as well as digital preservation of Islamic culture.]]>
