RP-Bert: An approach to detect and classify network intrusions based on a combination of Transfer learning and rules

We investigate the model generalization problem that exists for network traffic intrusion detection. Existing works have begun to apply some advanced algorithms and models from NLP domain to network security domain, but have some challenges. One of the biggest problem is that the amount of data for some existing network intrusions is not sufficient to train the current oversized models. For this purpose, we improve on ET-Bert and propose a method to detect and classify network intrusion based on the combination of migration learning and rules (RP-Bert). And we test and validate it on a large open source dataset BoT-loT, which Recall is achieved, Precision is achieved, F1 Score is achieved, Accuracy is achieved.