Towards Precision Agriculture Using Federated Learning-Driven Crop Recommendation System

This research paper proposes a federated learning (FL)-based crop recommendation system to support precision farming. The collaborative learning mechanisms are carried out by ensuring the privacy of FL participants is secured. In traditional crop recommendation systems, deep learning (DL) models were trained with a limited number of data points, resulting in lower model performance, especially when used in different environmental conditions. By utilizing this FL-based crop recommendation system, a continuous learning process can be guaranteed. Moreover, we also propose a DL model that can be used by both local participants and global servers to understand data distribution. This model is built using a Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM), combining these two networks and limiting the convolution and kernel sizes to deliver better performance compared to other state-of-the-art DL models. Based on extensive performance evaluations, the results show that the proposed CNN-LSTM model achieves 98.77% accuracy, indicating a 3.13% improvement in terms of accuracy compared to its counterparts.