Leveraging Federated Learning for Weather Classification in the Era of Smart Cities

The advent of smart cities has led to a significant increase in the collection and utilization of diverse data streams, including imagery datasets capturing various aspects of urban environments. This paper presents a novel approach to weather classification in the era of smart cities through the integration of federated learning (FL) techniques. This study aims to leverage decentralized computing to enhance the accuracy and efficiency of weather classification models. The federated model exhibits robustness and adaptability to the dynamic and heterogeneous nature of smart city environments. The findings of this research contribute to the advancement of weather monitoring and pre-diction capabilities in smart cities, fostering a more resilient and responsive urban infrastructure. Additionally, the federated learning approach demonstrated its effectiveness in managing decentralized datasets, ensuring the model training while main-taining data privacy in the context of smart city deployments. The study leverages the pre-trained ResNetSO architecture to achieve a commendable training accuracy rate of 87 % and a validation accuracy rate of 86 %. The implications of this research extend beyond weather classification, laying the groundwork for the broader integration of federated learning techniques in diverse applications within smart city ecosystems.