BalanceNet: Addressing Class Imbalance in AI-Powered Intrusion Detection Through Adaptive Sampling

Abstract

The constantly increasing cases of computer-attacks in the modern digitally connected world leader to the necessity of the most efficient intrusion detection systems (IDSs). Since innocuous traffic flow greatly outweighs the occurrence of attacks, one of the most crucial difficulties in intrusion detection systems is investigating the class imbalance of data flow from networks.   Since this is the case, it impacts the accuracy with which machine learning algorithms detect dangers to minority classes.   The research study introduces an intrusion detection system that uses adaptive sampling techniques to tackle the issue of network traffic class imbalance. It uses the UNSW-NB15 dataset, Extreme Gradient Boosting (XGBoost), and oversampling based on ADASYN, and it promises to improve the capacity to detect intrusions that impact minority classes. The model's 99.59% accuracy, 99.8% precision, 99.5% recall, and 99.6% F1-score indicate that it is very good at detecting harmful activity with few false alarms.   In comparison to LR, NB, and LSTM, XGBoost performs better across the board when it comes to critical metrics. The combination of adaptive data balancing with a robust ensemble classifier provides a scalable and robust solution to real-time network anomaly detection in complex and unbalanced network settings, which can be used to further develop intelligent cybersecurity systems.