Lack of ground truth for mobiflow_v2 dataset

[LWSen]

Hi, I have tried the autoencoder_v2 to evaluate the mobiflow_v2 dataset, but it seems that the labels of signaling messages are missing. Thus, it is difficult to identify whether the result is right. Here is the result:

[onehouwong]

Hi @LWSen. You can refer to here (https://github.com/5GSEC/MobiWatch/blob/main/src/ai/lstm_v2/test.py#L120) to see the ground truth (i.e., the abnormal messages in each attack trace)

[LWSen]

Hi @onehouwong , it seems that autoencoder_v2 and lstm_v2 perform worse than v1. Here is result: