Dynamic feature-based expansion of fuzzy sets in Neuro-Fuzzy for proactive malware detection
Andrii Shalaginov · 2017
Neuro-Fuzzy has been successfully applied in the malware detection from before. It gives flexibility in building an effective and human understandable rule-based detection model. Fuzzy variables consist of linguistic terms that are constructed based on the characteristics of the corresponding numerical features. This gives a level of abstraction that allows controlling the distribution drift and maintaining the appropriate performance when the number of terms is fixed. At this point the challenge can be seen when it comes to inclusion of a new term in a fuzzy set. Originally, this leads to a need of Neuro-Fuzzy model retraining. However, in case of Big Data retraining of the whole model may require enormous resources. In this paper we concentrate on mitigation of the retraining by means dynamic fusion of terms in fuzzy sets and corresponding rules selection. The algorithm Dynamically-Expanded Neuro-Fuzzy (DENF) was proposed to facilitate the different area of Information security such that pro-active malware detection or Intrusion Detection, where the properties of investigated data may change unpredictably.