Enhancing Time Series Analysis with GNN Graph Classification Models

The use of Graph Neural Networks (GNNs) in time series analysis is on the rise, yet the application of GNN Graph Classification in this field remains in its early stages. In our research, we repurpose GNN Graph Classification, traditionally rooted in disciplines like biology and chemistry, to delve into the intricacies of time series datasets. We demonstrate how graphs are constructed within individual time series and across multiple datasets, highlighting the versatility of GNN techniques beyond their standard applications. A key observation in our study was the sensitivity of the GNN Graph Classification model to graph topology. While initially seen as a potential concern for model robustness, this sensitivity turned out to be beneficial for pinpointing outliers. Our findings underscore the innovation of applying GNN Graph Classification to time series analysis, unlocking new dimensions in data interpretation. This research lays the groundwork for integrating these methodologies, indicating vast potential for their wider application and opening up promising avenues for future exploration.