Optimizing Cleanset Growth by Using Multi-Class Neural Networks

Abstract

Starting from 2005-2006 the number of malware samples had an exponential growth to a point where at the beginning of 2018 more than 800 million samples were known. With these changes, security vendors had to adjust - one solution being using machine learning algorithms for prediction. However, as the malware number grows so should the benign sample set (if one wants to have a reliable training and a proactive model). This paper presents some key aspects related to procedures and optimizations one needs to do in order to create a large cleanset (a collection of benign files) that can be used for machine learning training.