Deep Packet Filtering Mechanism for Secure Internetworks

Abstract

In this paper, we propose a Deep Packet Filtering Mechanism (DPFM) to analyze and filter malicious data packets moving between network environments. DPFM analyzes the behavior of malicious packets on the network and extracts information about the network as a sequence. After performing the word embedding process on the extracted sequence data using the word2vec technique, it detects malicious packets on the network by learning the LSTM model. In the past, research on filters to prevent malicious packets from entering the network by converting packets into data at the sending and receiving destinations and analyzing their purpose and maliciousness is insufficient. Since DPFM proceeds at the network boundary to analyze and extract malicious packets, primary detection is possible. In this paper, more accurate identification is possible by deep learning of network packets as well as OPcode and system calls, which are static analysis data.