Network intrusion detection system (NIDS) is a software system which plays an important role to protect network system and can be used to monitor network activities to detect different kinds of attacks from normal behavior in network traffics. A false alarm is one of the most identified problems in relation to the intrusion detection system which can be a limiting factor for the performance and accuracy of the intrusion detection system. The proposed system involves mining techniques at two sequential levels, which are: at the first level Naïve Bayes algorithm is used to detect abnormal activity from normal behavior. The second level is the multinomial logistic regression algorithm of which is used to classify abnormal activity into main four attack types in addition to a normal class. To evaluate the proposed system, the KDDCUP99 dataset of the intrusion detection system was used and K-fold cross-validation was performed. The experimental results show that the performance of the proposed system is improved with less false alarm rate.
Supervised contrastive learning over prototype-label embeddings for network intrusion detection
