Intrusion in Wireless Networks Detection Using Machine Learning
DescriptionAs people are increasingly using wireless devices, it is important to build a reliable detection system that can detect intrusions in a wireless network. Recently, machine learning methods have been successfully applied to detect intrusions in wireless networks. These machine learning models are built on a set of training data and can identify most future attacks in the network. Detecting intrusion in a wireless network using machine learning is a challenging problem because the number of intrusion-related data points is much smaller relative to the number of normal network transfers data points.
For this project, we investigated which machine learning model performs best to detect intrusions in a wireless network. Several tree-based methods: XGboost, Catboost, LightGBM, and Random Forest are trained on the Aegean Wi-Fi Intrusion Dataset (AWID), a publicly available data with records of wireless network features during intrusion and normal processing. By passing the test dataset to the above-mentioned machine learning models, we evaluate these models for accuracy, precision, recall, and time of execution. Cleaning the AWID dataset and selecting the right features from it is an important step to ensure our models are making the best predictions. Therefore, after running the above-mentioned machine learning models, we use SHAP (SHapley Additive exPlanations) to find the features of the AWID dataset that contribute most to those machine learning models while detecting intrusions. Identifying those important features will allow us not only to get rid of insignificant attributes during future experiments but also to better understand how the models work.