A Survey of Feature Extraction and Feature Selection Techniques used in Machine Learning-Based Botnet Detection Schemes

Abstract

Machine learning techniques have been widely used for the classification of botnets as they have been argued to have improved strengths compared to signature-based approaches. The level of performances of some of these detection schemes have been traced to the relevance of the features used for the classification models. Therefore, extraction and selection of the most discriminative features in the classification of botnets is an important research area. It has equally been found that when a Machine-learning based approach is being used to identify botnets, the dataset chosen has to be real and representative. Feature extraction and Feature Selection are necessary steps prior to using a Machine Learning-based classification algorithm for identifying botnets. The reason for the pre-processing and feature selection steps in Machine Learning-based model is to be able to remove irrelevant and redundant data in the experimental dataset, minimize computational complexity, and increase both model simplicity as well as accuracy. This paper provided a survey of various feature extraction and feature selection methods that have been used by researchers that proposed Machine-Learning based botnet detection models. The main purpose of this approach is to provide a better understanding and insights on how improved botnet detection mechanisms can be achieved through enhanced feature extraction and selection methods.