Novel seismic neural network approach for robust intrusion detection using texture-based features

As digital networks become increasingly popular, cybersecurity is also becoming a more and more heavily noticed issue. It requires strong methods to detect problems and remedy threats. In this paper, we point out the built-in limitations of traditional Intrusion Detection Systems (IDS), especially their dependence on assumed rules and the high false positive rate. However, it acknowledges that cyberattacks are dynamic in nature; hence the systems should have the capability to detect known and unknown threats. The research proposed in this research leverages Machine Learning (ML) and Deep Learning (DL) techniques to improve precision and adaptation of Network Intrusion Detection Systems (NIDS). It shows how advanced architectures like the Convoluted Neural Networks (CNNs), Long Short-term Memory (LSTM) and autoencoders can do better for anomaly detection. The developed methodology includes classification in which texture features are extracted, normalized and a Seismic Neural Network (SNN) is utilized. Our experiments on NSL-KDD and UNSW-NB15 datasets show these improvements; with improved accuracy, precision, re-call, and F1-scores of 99.67% and 98.34% compared to 99.04% and 61.73%, respectively. Our results achieve over 3% absolute improvement in accuracy and significantly lower false positive rates than state of the art technique. Furthermore, this approach can detect previously undetected or rare threats and proves to be robust across different attack classes. The results represent an important step in the design of IDS. It covers the existing gaps and extend the research to real time Intrusion Detection (ID) mechanisms.