Sine-Cosine Zebra Optimization Algorithm with Mixture of Experts based Long Short-Term Memory for the Detection and Classification of Cyber Treats

The use of internet application is rising rapidly where the number of users is increasing day by day creates traffic in the network which led path for the growth of cybercrimes. To avoid the cyber treats various advance technologies have been introduced for the detection of unauthorised users in the network. The existing methods tried to improve the accuracy of cyber threat section model using various algorithms but did not reach the target due to the issues such as imbalanced data, variable constraints and overfitting. The paper proposed Sine-Cosine Zebra Optimization Algorithm with Mixture of Experts-Long Short-Term Memory (SCZOA-MoELSTM) for the detection and classification of cyber treats. The min-max normalization rescaled the ranges of the data. The Sine-Cosine Zebra Optimization Algorithm (SCZOA) was used for the selection of optimal features. Mixture of Experts-Long Short-Term Memory (MoELSTM) captured the long-range values of the features that enhanced the performance of cyber threat detection and classification. The proposed SCZOA-MoELSTM model obtained accuracy of 99.04% and 96.85%, precision of 98.22% and 95.73%, Detection Rate (Recall) of 98.98% and 95.67%, F1-Score of 98.59% and 95.69% respectively for CICIDS2017 and UNSW-NB15 datasets compared to the existing Long Short-Term Memory (LSTM).