Predicting Pest Infestation Patterns with Graph Convolutional Networks (GCN) in Precision Farming

Due to the increases pest impacts on agricultural productivity, effective pest prediction models are required for efficient pest control in precision farming. The framework outlined in this research uses a Graph Convolutional Network (GCN) to estimate pest infestation by exploiting spatial and temporal relationships present in agri-ecosystems. The methodology combines weather, soil, pest, and crop health data into a graph-based representation of spatial units, and the directed connections between them that reflect their proximity and climatic similarity. The GCN converts these graphs to learn such complex relationship as the infestation prediction at various severity levels. The proposed model has produced a better accuracy of 94.8%, better precision of 95.2%, better recall of 94.5% and better F1-score of 94.8% as compared from Random forest, SVM, LSTM, CNN as per the comparison of table 3. Comparing them focuses on graph construction approaches and temporal aspects, mixed P/S and temporal emb has improved the model performance. This framework is practical for further pest management and employing the result without harming the environment. The results show the effectiveness of GCNs for use in pest prediction in precision farming and the opportunity to develop sustainable and scalable data-based agriculture.