Rice cultivar clustering using federated K-means: focusing on advancing agriculture 4.0 applications

The clustering of rice cultivars based on agricultural attributes is essential for understanding diversity, optimizing cultivation, and enhancing yield. However, the decentralized nature of agricultural data poses privacy challenges. This study introduces the flexible federated K-means (FlexFed-KM) framework, a privacy-preserving solution that enables collaborative clustering across institutions without sharing raw data. FlexFed-KM integrates federated averaging (FedAvg), federated proximal (FedProx), and quantized federated averaging (QFedAvg) for enhanced clustering flexibility and adaptability. In experiments with 130 rice cultivars sourced from the Indian Council of Agricultural Research (ICAR), QFedAvg demonstrated the highest clustering performance, achieving a silhouette score of 0.402 and an inertia value of 22.64. FedProx also improved clustering accuracy over FedAvg, with a silhouette score of 0.382 and inertia of 24.91. These findings underscore FlexFed-KM’s potential to support Agriculture 4.0 applications by enabling secure, scalable analysis of distributed agricultural data.