Enhancing Wi-Fi Network Security Through Stacked Machine Learning Models

The paper presents a wireless intrusion detection system (WIDS) based on model stacking for IEEE 802.11 Wireless Local Area Networks (WLANs). The proposed system utilizes model stacking, a machine learning (ML) technique in which the predictions of several ML models, called base classifiers, are combined to improve the overall performance of the system. Various stacking algorithms and ML architectures, including recurrent neural networks (RNN) and Convolutional Neural Networks (CNN), are implemented and compared. For the training of ML models and the evaluation of the proposed solution, the Aegean Wi-Fi Intrusion Dataset 3 (AWID3) was used. Moreover, the feature selection and data preprocessing are described in detail. The model stacking algorithms were shown to improve the WIDS performance in multiple scenarios, with the best ones achieving very high 99.878% accuracy and 99.507% F1 score for binary and 99.858% accuracy and 99.616% F1 score for multiclass classification.Theexperimental results demonstrate that selected stacking configurations achieve improved detection performance compared with the individual base classifiers while highlighting the influence of the chosen aggregation strategy. Most notably, the implementation of model stacking is shown to greatly reduce the number of false positives in most of the tested scenarios, which is a pressing issue in most intrusion detection systems (IDSs).