Intrusion Detection based on Incremental Learning
Islem Chouchen, Farah Jemili · 2023
Machine learning and deep learning have become essential in enhancing the performance of intrusion detection systems. While existing research on intrusion detection systems utilizing data mining and machine learning has shown effectiveness, it typically involves training static batch classifiers that identify intrusions without considering the time-varying characteristics of the regular data stream. This paper aims to propose an ensemble adaptive approach for online intrusion detection using stream-oriented learning, which can effectively adapt to concept drift in real-world environments. The technique involves the utilization of an ensemble Adaptive Random Forest classifier and Support Vector Regression (SVR) with the ADWIN change detector to detect changes in a data stream and dynamically adjust to drift detection in the streamed data. This approach enables agile adaptation against unknown intrusions and eliminates the need for frequent model retraining over time.