Use of AI in Operational Technology Networks and Packet-Based Attacks Detection
Zoltán Dobrády, Szilárd L. Takács, Timót Hidvégi · 2023
This research is focused on cybersecurity, including the detection of packet-based attacks. We collected a large amount of data by creating Honeypots and hosting them on virtualised private servers (VPS) with open IP addresses. The acquired data was analysed using different deep learning methods, such as Long Short-Term Memory (LSTM) and one-dimensional convolutional neural network techniques. These algorithms were used to compare the measurements with currently used packet analysis techniques, resulting in the identification and development of the most efficient packet analysis procedure. Additionally, we conducted regression tests in isolated and simulated environments using the attack mechanisms that had already been detected. Once the packet analysis concept was developed, our goal was to improve a classification algorithm. The construction of a penalty decision algorithm was crucial. We also conducted extensive regression testing of the concept from various perspectives. Upon completion of our investigation, it was discovered that natural and statistically-based language models can identify cyber-attacks. Statistical models that better fitted were SVC, Logistic, and Naive Bayes, with a 69 % accuracy for packet-based attack detection.