Real-Time Intrusion Detection Using Adaptive Sliding Windows for Concept Drift

Real-time intrusion detection has become crucial as a result of the rapid development of networks. This is because the distribution of the data and behaviors changes over time, a phenomenon known as concept drift. In order to address the data drift, this study suggests that an online adaptive sliding windowing method is used to tackle the concept drift, which gives a timely response for incoming data packets. Initially, collected different Intrusion Detection System (IDS) datasets like NSL KDD from the concerned repository and dataset evaluated by conventional machine learning models such as Logistic Regression(LR), Random Forest (RF), Decision Tree(DT), K-Nearest Neighbor (KNN), Gradient Boosting classifiers (GBT), and Light(GBM) and the results showed low detection rate 70.24%, 76.77%, 76.83%,77.20%,78.02% and 79.35% due to training and testing datasets consists of unequal class distribution and concept drift. Applying the oversampling, Undersampling, and SMOTE approaches, the accuracy was somewhat improved to 79.77% and 81.21% while using Synthetic Minority oversampling techniques (SMOTE) to address the majority and minority issues known as data imbalance. To further enhance the detection rate, developed a proposed model called Adaptive Drift-aware Windowing Intrusion Detection System with Optimization (ADWISE) was developed, combining adaptive sliding windows with random search hyperparameter tuning optimization. The proposed ADWISE framework achieves a top accuracy of 98.27% while effectively managing both class imbalance and concept drift.