Federated learning methods for collaborative multiclass classification of dry beans

Federated learning (FL), a decentralized approach, enables multiple institutions to collaboratively train a shared model while keeping their data localized. This ensures data privacy and security, crucial for sensitive agricultural datasets. This paper presents a comprehensive study on the application of FL methods for the collaborative multiclass classification of dry beans. We introduce the FL assisted agriculture 4.0 framework, which utilizes federated average and federated proximal aggregation techniques to create a consolidated model. Within this framework, a common multilayer perceptron model is trained to analyze decentralized agricultural data. For this study, we utilize the dry bean dataset from Kaggle for classification within a FL setup. The study evaluates the performance of the aforementioned FL algorithms in the context of multiclass classification, comparing them against traditional centralized models. To test the robustness of FL techniques on horizontally distributed agricultural data, we categorize the dataset into two types: independent and identically distributed (IID) and non-IID. In our simulation involving 10 clients, we explored scenarios featuring 0%, 40%, and 80% stragglers-nodes encountering delays. The results indicate that federated proximal outperformed federated average in handling delays and enhancing efficiency in distributed learning. Experimental results demonstrate that FL can achieve comparable, and in some cases superior, classification accuracy with significant improvements in data privacy. The implications of this study suggest a promising future for FL in agricultural applications, offering a pathway to more secure and collaborative machine learning frameworks.