Multi-Classification of Heritage Buildings using Federated Learning CNN: A Comparative Analysis of Client-Side and Global Model Performance

The study suggested a multi-classification method using a federated convolutional neural network (CNN) for five distinct categories of historic buildings. Due to the federated design, data privacy was maintained while training on data spread across numerous clients. Various measures, including accuracy, precision, recall, and F1 score, were used to thoroughly evaluate the model's performance on both an individual client and an aggregated global model level. The results showed the federated CNN to be very successful, with F1 ratings of 0.87, 0.89, and 0.84 for different categories of historic buildings. High precision and recall metrics further supported the model's ability to classify landmark structures accurately. The research supports the usage of federated CNNs for activities that call for high performance and data privacy overall.