Brief

Researchers have developed a novel method to enhance Graph Neural Networks (GNNs) for dust source emission forecasting by incorporating proximity graphs. These graphs, including Delaunay triangulation, Gabriel graph, k-Nearest Neighbor graph, and Yao graph, are used as input for various GNN architectures like GraphSAGE, Graph Convolutional Networks, and Graph Attention Networks. The study demonstrates that GNNs utilizing proximity graphs significantly outperform those using random graphs and are also superior to traditional Long Short-Term Memory (LSTM) models in accurately predicting dust storm phenomena. AI

This paper introduces a novel approach to binary classifiers by incorporating per-class regularization hyperparameters within Gabriel graph-based systems. The method enhances outlier elimination and addresses class imbalance by allowing flexible thresholds for majority and minority classes. Experimental results using the Friedman test indicate that this adaptive regularization improves classifier performance. AI