Since feature selection in artificial intelligence refers to how researchers can target the most relevant features of a given model in order to iterate over those features more intelligently to construct a better future model, feature popularity is a more advanced heuristic than compares and contrasts features of different sub-models in order to garner a more accurate model (though it’s obviously a lot more complicated and technical than that). Feature popularity has never been previously applied to machine learning and data mining, which is unfortunate because such an application has the potential to lead to great advances in data-based cybersecurity. Luckily, researchers in this paper did just this, applying feature popularity to machine learning based off of “ensemble feature selection techniques or FSTs”. Researchers found that when running these heuristics on several web-attack datasets, they were able to understand more about the commonalities of different classes of cyberattacks and have more easy to understand models about the nature of cyberattacks. Using these advances, security professionals and researchers will be able to more deftly prevent cyberattacks in the future.
Author:
Richard Zuech, John Hancock & Taghi M. Khoshgoftaar
