Anomaly Detection Method Based on GNN

Anomaly detection is crucial in multiple fields, as it can quickly identify data errors and missing data, while traditional methods are limited by single factor identification. Faced with the complexity of big data in the digital age, especially the high-dimensional nature of time series data, traditional methods have become ineffective due to poor computational scalability and curse of dimensionality. To this end, a novel multidimensional sequence anomaly detection model based on optimized graph neural network is introduced, which adopts the graph bias network (GDN) framework as its core and innovatively integrates the optimization prediction function of Transformer model. This model significantly improves the accuracy of anomaly detection by accurately predicting the future values of adjacent time series and deeply mining the dependencies and features between data. Experimental verification shows that this method exhibits higher accuracy and computational efficiency on multiple standard datasets.