Viewing Historythrough the Lens of Artificial Intelligence: Classification of late Ottoman and early Republican period buildings in Türkiye with Convolutional Neural Network(CNN)

Abstract

This study employs Convolutional Neural Networks (CNNs) to classify late Ottoman and early Republican period buildings in Türkiye, offering a unique lens through artificial intelligence (AI) to examine architectural styles. By training on a specially curated dataset, including images of 16 architects’ works, the study achieves accuracy rates of 84.65% for a limited architect dataset and 74.08% for the full architect dataset. EfficientNet emerges as the optimal architecture, surpassing Baseline, VGG, and ResNet models. Through t-Distributed Stochastic Neighbor Embedding (t-SNE), the model visualizes relationships among architects' styles. This research not only provides a new perspective on Turkey's architectural heritage but also establishes a platform for future AI-driven architectural analyses and design paradigms. © 2024 Elsevier B.V., All rights reserved.