Comparison of recurrent and convolutional neural networks for analyzing and predicting cyber security threats

Abstract

Cyberattacks are becoming increasingly complex and sophisticated, which requires the development of new methods for analyzing and predicting threats. Traditional methods such as signature analysis and statistical models are often ineffective against advanced threats. Recurrent neural networks (RNN) represent a promising tool for this task due to their ability to process time-series data and identify complex patterns. Today's information systems face an ever-increasing number of cyber threats. These threats range from simple phishing attacks to sophisticated campaigns carried out by organized criminal groups. Securing information resources requires the development of effective methods to analyze and anticipate potential threats. Current approaches include the use of machine learning and artificial intelligence (AI). In order to use AI as part of information security challenges, particularly cybersecurity, it is necessary to understand which algorithm is best suited for analyzing, predicting and detecting cyber threats in modern and complex information systems. This paper discusses the potential applications of RNN for analyzing and predicting cyber threats and provides examples of successful implementations of these models in the cybersecurity domain. A database consisting of 40000 cyberattacks was used to train the model. The purpose of this paper is to investigate the potential applications of recurrent neural networks in the context of analyzing and predicting information security threats. The scientific novelty of this article lies in the comparison of the results of different neural networks. Research methods: system analysis of existing machine learning methods, theoretical formalization, experimentation.