Identifying Traditional Malay Building Architectural Styles Using Vision Transformer Architecture

Abstract

The preservation and documentation of traditional Malay buildings is a significant challenge, especially in identifying diverse architectural styles, which is often done manually. This study aims to optimize digital architecture using Vision Transformer (ViT) for identifying Malay architectural styles, such as Riau Malay and Pontianak Malay, by measuring model performance using Precision, Recall, and F1-Score. The method used is ViT-based deep learning trained using a dataset of traditional building images. The data was divided using an 80:20 and 70:30 ratio for training and testing data. The model was optimized to improve accuracy and prevent overfitting using regularization techniques. Testing results show that the ViT model achieved excellent Precision, Recall, and F1-Score values, with Precision and Recall reaching 0.99 on the training data, and 0.98 for Riau Malay House Types and 0.97 for Pontianak Malay Traditional Houses on the test data. This proves that ViT can automatically and accurately identify Malay architectural styles. This research has the potential to be applied in digital preservation, traditional building documentation, and the development of AI-based applications for the cultural and tourism sectors.