Raja Nandini, Kalyana Chakravarthi Agnihothram and S. Vinod Kumar |
Computer Science and Engg., Malla Reddy Engineering College for Women, Hyderabad, Telangana, India.
Abstract
An intrusion detection system aims to stop harmful attacks. Furthermore, attackers’ strategies and tools are always evolving. In our most recent work, we suggested using support vector machines (SVM) and random forests in military defense settings (KDD dataset). Regardless of whether SVM and RF have shown good accuracy and precision, they failed to lower FN rates, and other deep learning modalities have been examined. Several experiments were carried out on the KDD dataset in an attempt to boost efficiency and decrease FN rates. The hybrid SVM+LSTM model showed improved accuracy, recall, and precedence along with a successful decrease in
FN rate. This paper uses benchmark datasets, including KDD, NSL-KDD, CICIDS2017, and UNSW-NB15, to present a synergistic hybrid model for proactive intrusion detection in cyber-physical networks. After evaluating several models and techniques, such as CNN, RNN, autoencoder, gradient boosting, decision tree, and K-means, the hybrid SVM+LSTM model outperformed the others in terms of lowering the FN rate and improving detection efficiency.
Keywords: Autoencoder, CNN, CICIDS2017, Deep Belief Network (DBN), Gradient boosting, IDS, K-Means, KDD, ML, RNN, NSL-KDD, SVM+LSTM, Transformer models, UNSW-NB15, XGBoost.
View PDF