Machine Learning predictive model to detect impersonation attacks
A predictive model (i.e., a machine learning classifier) capable of distinguishing between “intrusive” traffic, called threats or attacks, and “good” normal traffic (i.e., binary classification).
Uses the impersonation attacks part of the Aegean WiFi Intrusion/threat Dataset (AWID2), https://icsdweb.aegean.gr/awid/awid2
A Variational Autoencoder is used to generate 20 additional features to add to the data set.
Filter methods are used to select the best features, generated features are included.
A selection of algorithms are spot checked using K fold cross validation to find the most suitable for the data
Hyperparameter tuning is performed to tune the selected models.
A variety of evaluation metrics are used to assess model performance for comparison to other benchmark studies. A Naive Bayes model is selected as the best performing model for the task.