CNN-BiLSTM Attention Model with Bagging Adaboost Voting for Intrusion Detection

Abstract

Network intrusion detection is an essential component of cybersecurity, and it requires robust and clever methods for detection and prevention of malicious activities. This research explores an improved deep learning technique which attempts to enhance the accuracy of attack detection using the NSL-KDD, UNSWNB15 and CIC DDoS2019 datasets. To find spatial and sequential dependencies in network traffic data, a wide range of machine learning and deep learning methods are used. These include CNN structure and RNN structure. The proposed approach fuses sequential pattern recognition, deep feature extraction and attention technique to achieve better detection accuracy by paying attention to the interesting attack characteristics. Using experiments, we found that the Voting Classifier (BaggingDT + BagRF) works the best. It achieved 98.1% accuracy in the CIC-DDoS 2019 dataset, 99.9% accuracy in the NSL-KDD dataset and 92.9% accuracy in the UNSW-NB15 dataset. The results in these examples demonstrate that ensemble learning is a suitable strategy for boosting accuracy, memory and generalisation of intrusion detection on a broad spectrum of patterns. The results demonstrate that the integration of multiple classifiers and enhanced representation of features significantly boosts the network intrusion detection capability, positioning it as a scalable and robust solution for practical cybersecurity challenges.