|
Abstract
|
Classification is a core statistical concept and is the process of assigning a given observation to one of a set of categories based on its characteristics. So far, various types of classification algorithms, such as logistic regression, decision trees, support vector machines, neural networks, and graph neural networks have been introduced to classify observations by analyzing their properties. This paper focuses, however, on classifying variables. In other words, we aim to propose a method leveraging the R package, BDgraph, and a graph neural network called Graph attention network (GAT) to classify
variables in a given graphical model. The BDgraph package facilitates the generation of graphical models based on Bayesian structure learning on a multivariate dataset. Additionally, we incorporate the concept of Graph Attention Network, GAT, and introduce our novel architecture as BGAT to utilize it for variable classification of graphical models. We evaluate its performance through different simulation studies and a real-world dataset. Furthermore, we conduct some experiments that confirm our proposed architecture outperforms the state-of-the-art approach, GAT.
|