Unwinding the Significance of Federated Learning for Smart Agriculture

Smart agriculture leverages cutting-edge information and communication technologies to optimize agricultural practices, and machine learning plays a significant role in achieving this goal. However, due to the conventional nature of centralized learning, machine learning-oriented solutions in agriculture face significant challenges, primarily the network overburden, time-delayed response and limited computational capabilities of edge nodes. This leads to the requirement of a more efficient and decentralized approach for learning, which can distribute the computational load and enable real-time data processing closer to the data sources. The presented paper un-winds the significance of Federated Learning in addressing these challenges associated with Centralized Learning, particularly in agriculture. The paper explores Federated Learning's ability to leverage the collective intelligence of distributed data sources while maintaining computational efficiency and scalability. It presents it as an ideal solution for the agricultural sector, where data is often fragmented and collected in diverse environments. In this work, a methodology is adopted to compare and analyze the performance of Federated Learning with Centralized Learning using classification-based statistical parameters viz. Accuracy, Precision, Recall and F1-Score and hosting nodes' parameters viz. Memory usage, CPU utilization, Model Converging Time and Node Energy Consumption. The experimental results clearly depict the edge of Federated Learning in analyzing the crop recommendation, as compared to the Centralized Learning.