Crop Irrigation Advisory System Using Federated Logistic Regression

Federated learning (FL) is a collaborative learning algorithm that builds ML models using knowledge from distributed clients at various locations. Federated learning is valuable when data is collected from heterogeneous environments, and a unified ML model needs to be built with minimal sharing of data. In this study, a federated irrigation advisory system is developed for an agricultural application that helps the farmer to water their field based on agricultural field parameters. The application of machine learning techniques in the agricultural domain is still in its infancy. Agricultural farms employ different sensors and have natural variations in soil type and environmental conditions. A federated model framework is suitable for the development of an irrigation advisory system as heterogeneous data is gathered using a variety of sensors spread across several farms. FL-based Logistic Regression is designed to predict whether to water the field or not based on the field parameters: temperature, humidity, soil moisture content, number of days since planting the crop, and crop type. Federated learning algorithms are designed with a client-server architecture. Flower, an open-source federated framework, which used to build and test the irrigation advisory application. Two clients and a server are used to build the FL-based Logistic Regression model by exchange of model parameters. Weights and bias of the logistic regression model are aggregated by the server from multiple clients. These aggregated model parameters are sent to clients to build the regression model with its own data. Experiments were conducted to understand the performance of FL in learning the model parameters for prediction. The model’s performance is evaluated using metrics to understand the converge patterns and parameters that affect FL.