Machine Learning for Web Intrusion Detection: A Comparative Analysis of Feature Selection Methods mRMR and PFI

摘要

Select from the best features in a complex dataset that is a critical task for machine learning algorithms. This work presents a comparative analysis between two resource selection techniques: Minimum Redundancy Maximum Relevance (mRMR) and Permutation Feature Important (PFI). The application of PFI to the dataset in issue is unusual. The dataset used in the experiments is HTTP CSIC 2010, which shows great results with the mRMR observed in a related work [22]. Our PFI tests resulted in a selection of features best suited for machine learning methods and the best results for an accuracy of 97% with logistic regression and Bayes Point Machine, 98% with Support Vector Machine, and 99.9% using an artificial neural network.