Graph Neural Network Classification Algorithm

Graph neural network is a machine learning method that aims to learn features through graph-based structures so as to capture relational patterns efficiently,and is highly favored by academia and has become a popular research content among scholars,showing excellent results in applications in industrial engineering,pattern recognition,finance,and even medical fields.However,the traditional graph neural network has some drawbacks.The traditional model is constructed by aggregating the information of neighboring nodes of each node and iterating continuously,which will produce the drawbacks of deep layers and high time complexity.In this paper,we propose a new method and idea for the construction of graph neural network model,which not only has a simple algorithm,but also the classification accuracy has been improved.The main innovation points of the paper are as follows:First,a localization-based graph neural network classification model is proposed.The model has the following outstanding advantages: a change from the traditional graph neural network method of aggregating the information of neighbor nodes of each node,so that the model reduces the number of iterations and speeds up the computational efficiency.The adoption of Bayesian algorithm into the construction of the graph neural network model not only alleviates the assumption of conditional independence of Bayesian algorithm,but also provides a new idea to the construction of graph nodes in the graph neural network model,which is an extension of the weighted Bayesian algorithm and a bold integration of Bayesian algorithm and graph neural network.The set of edges in the graph structure is calculated by Euclidean distance,in which the data can exist in a multidimensional space with unrestricted dimensionality and more intuitive.In addition,a particle swarm optimization algorithm is used to optimize the graph neural network model by projecting the parameters from the low-dimensional space to the high-dimensional space,which results in a fast convergence rate and improves the accuracy of the model.Second,proposed global-based graph neural network classification model.The model is optimized and improved relative to the previous section for the model construction part to improve the accuracy of the model.In this part,we use the Marcian distance added to the operation of the graph structure edge set,compared with the Euclidean distance,the Marcian distance takes into account the connection between various characteristics,which is not affected by the original data measurement unit and can exclude the interference of correlation between variables,and the calculation of the Marcian distance is based on the overall sample,which is more correlated with the overall sample.Third,graph neural network classification models were applied to atrial fibrillation heart disease identification diagnosis and breast cancer disease diagnosis.A local-based graph neural network classification model was used for atrial fibrillation diagnosis,and a global-based graph neural network classification model was used for breast cancer diagnosis.