Comparative Analysis of Word Vectoring Methods for Malware Detection

Abstract

The development of mobile applications and platforms has triggered the development momentum of applications that aim to harm. While traditional malware detection studies are ineffective against these modern applications, machine learning methods have also created a new study area. However, since these methods have been developed for naturally structured objects and/or texts also cause certain performance deficiencies in the studies. In this study, the performances of the word vector methods developed for a natural language such as TF, TF-IDF, which work statistically, and Fasttext, SkipGram, and GloVe methods, which are designed to establish semantic and syntactic relationships in the text, on malware detection were compared.