Impact of Federated Learning in Agriculture 4.0: Collaborative Analysis of Distributed Agricultural Data
Ankush Kumar Gaur, Joseph Arul Valan · 2024
Agriculture 4.0 integrates traditional farming methods with advanced technologies, yet its adoption faces significant challenges. This article identifies these obstacles and proposes leveraging Federated Learning (FL) to overcome them. However, the application of federating machine learning algorithms within the agricultural domain is still in its infancy. We introduce the Federated Learning Assisted Agriculture 4.0 (AgriFL 4.0) framework, designed to ensure privacy while collaboratively analyzing distributed agricultural data. Within AgriFL 4.0, a common Multilayer Perceptron (MLP) model is trained locally to analyze decentralized agricultural data. Additionally, a consolidated model is formed at the server using the Federated Average aggregation technique. For this study, we employ the Dry Bean Dataset from Kaggle, assembled in 2020, consisting of 13,611 instances with 16 attributes, including dimensions and shape forms crucial for seed classification. To evaluate AgriFL 4.0's performance on horizontally distributed agricultural data, we categorize the dataset into two types: lID and non-lID. In our simulation involving 10 clients, we explored scenarios featuring 0%, 40%, and 80% stragglers-nodes encountering delays. Moreover, we compare the federated and non-federated model performances. The results indicate that the framework works well with lID data, and the federated model outperforms the non-federated model in terms of classification accuracy and other performance metrics. However, with non-lID data, the framework's performance degrades as the count of stragglers increases. This underscores the critical role of federated learning in advancing Agriculture 4.0.