Classification of network security attacks using novel enhanced restricted Boltzmann machine-based bitterling fish optimization algorithm

In the past, intrusion detection was mostly dependent on established techniques like protocol control, anti-virus software models, firewalls, and the encryption-decryption approach. Although these techniques were effective in detecting a small count of attacks, they had trouble detecting many attacks and had a significant false positive rate. But when harmful attacks get larger, new problems arise that require an all-encompassing reaction. Hence, this article plans to perform network security attack classification using novel intelligent deep learning techniques. Initially, the data is gathered from the standard benchmark datasets, such as the CICIDS 2017 and CSE-CICIDS 2018 datasets. The collected data undergoes pre-processing using the Min-Max normalization method. Next, the features are extracted from these pre-processed data using the Principal Component Analysis (PCA) method. These extracted features undergo the final classification stage using the novel Enhanced Restricted Boltzmann Machine (ERBM), in which the parameter tuning of RBM is accomplished by the nature inspired optimization algorithm referred as Bitterling Fish Optimization (BFO) algorithm with the intention of accuracy maximization as the major fitness function. This novel ERBM-BFO classifies the final output into various attacks, such as normal, DoS attack, DDoS attack, brute force attack, botnet attack, web attack, infiltration attack, and port scan attack, respectively. The outcomes suggested that the method is effective and suitable for identifying attacks in contemporary situations. The proposed ERBM-BFO method for the network security classification model in terms of accuracy is 9.35%, 6.85%, 4.47%, and 2.19% better than LSTM + AE, CNN+LSTM, MLP, and RBM, respectively. Similarly, the proposed ERBM-BFO method for the network security classification model with respect to recall is 9.44%, 6.92%, 4.51%, and 2.20% advanced than LSTM + AE, CNN+LSTM, MLP, and RBM, respectively.