Fusion of Feature Selection Techniques and Machine learning Algorithms for Attack Classification on 802.11 Wi-Fi AWID Dataset

Vulnerabilities pertaining to the Media Access Control (MAC) layer are the primary reason for the existence of attacks in 802.11 Wi-Fi networks. In this paper, we aim to explore the Aegean Wi-Fi intrusion dataset (AWID) and design a model for detection and classification of network intrusions. There are three classes of attacks present in the dataset: impersonation, flooding, and injection. In intrusion detection systems (IDS), machine learning classifiers are one of the most effective solutions for preventing unauthorised access to resources. The key aspect of building a model based on machine learning is feature selection. In this paper, we have used four machine learning classifiers, namely Decision Tree, Random Forest, Gaussian NB, and Bernoulli Naive Bayes, for the classification of data as normal or intrusive. We have further applied filter methods and wrapper methods for feature selection. To the best of our knowledge, machine learning along with feature selection has never been used on the AWID-CLS-R dataset. The proposed work achieved the highest accuracy of 95.2% using a random forest classifier in 104 seconds using filter methods as the feature selection technique. In addition, we have used a confusion matrix and model building time for evaluating other machine-learning classifiers.