I am working on an anomaly detection problem using a graph neural network. However, I am not sure which will be the best way to convert my CSV data into graph data. I have 115 different attributes for each timestamp. You can find the example dataset here; Example dataset for anomaly detection in IoT devices. This is only for PyTorch Discussion forum. · GitHub
I want to test each time stamp for an anomaly. Which will be the best way to convert CSV data into graph data?
I was thinking of considering each attribute as a node and connecting all of them with each other by the undirected edge (since I don’t know their connection to each other). Then I will have a single feature for each node and a lot of edges. To be specific
feature_shape == [115, 1] and
edge_shape == [2, 6555]. But I am not sure if this is a good way to model it, and what will be the best method to work on this on node embedding or graph embedding?
PS: I am using an autoencoder model to detect anomalies.
Thanks in advance. Hope I am clear about my question