As cyber-attacks increase in unencrypted communication environments such as the traditional Internet, protected communication channels based on cryptographic protocols, such as transport layer security (TLS), have been introduced to the Internet. Accordingly, attackers have been carrying out cyber-attacks by hiding themselves in protected communication channels. However, the nature of channels protected by cryptographic protocols makes it difficult to distinguish between normal and malicious network traffic behaviors. This means that traditional anomaly detection models with features from packets extracted a deep packet inspection (DPI) have been neutralized. Recently, studies on anomaly detection using artificial intelligence (AI) and statistical characteristics of traffic have been proposed as an alternative. In this review, we provide a systematic review for AI-based anomaly detection techniques over encrypted traffic. We set several research questions on the review topic and collected research according to eligibility criteria. Through the screening process and quality assessment, 30 research articles were selected with high suitability to be included in the review from the collected literature. We reviewed the selected research in terms of dataset, feature extraction, feature selection, preprocessing, anomaly detection algorithm, and performance indicators. As a result of the literature review, it was confirmed that various techniques used for AI-based anomaly detection over encrypted traffic were used. Some techniques are similar to those used for AI-based anomaly detection over unencrypted traffic, but some technologies are different from those used for unencrypted traffic.

Cyber threats to industrial control systems (ICSs) have increased as information and communications technology (ICT) has been incorporated. In response to these cyber threats, we are implementing a range of security equipment and specialized training programs. Anomaly data stemming from cyber-attacks are crucial for effectively testing security equipment and conducting cyber training exercises. However, securing anomaly data in an ICS environment requires a lot of effort. For this reason, we propose a method for generating anomaly data that reflects cyber-attack characteristics. This method uses systematic sampling and linear regression models in an ICS environment to generate anomaly data reflecting cyber-attack characteristics based on benign data. The method uses statistical analysis to identify features indicative of cyber-attack characteristics and alters their values from benign data through systematic sampling. The transformed data are then used to train a linear regression model. The linear regression model can predict features because it has learned the linear relationships between data features. This experiment used ICS_PCAPS data generated based on Modbus, frequently used in ICS. In this experiment, more than 50,000 new anomaly data pieces were generated. As a result of using some of the new anomaly data generated as training data for the existing model, no significant performance degradation occurred. Additionally, comparing some of the new anomaly data with the original benign and attack data using kernel density estimation confirmed that the new anomaly data pattern was changing from benign data to attack data. In this way, anomaly data that partially reflect the pattern of the attack data were created. The proposed method generates anomaly data like cyber-attack data quickly and logically, free from the constraints of cost, time, and original cyber-attack data required in existing research.

Anomaly detection systems are being studied to detect cyberattacks in industrial control systems (ICSs). Existing ICS anomaly detection systems monitor network packets or operational data. However, these anomaly detection systems cannot detect control logic targeted attacks such as Stuxnet. Control logic tampering detection studies also exist, but they detect code modifications rather than determining whether the logic is normal. These tampering detection methods classify control logic as abnormal if any code modifications occur, even if the logic represents normal behavior. For this reason, this paper proposes an anomaly detection method that considers the structure of control logic. The proposed embedding method performs embedding based on control logic Instruction List (IL) code. The opcode and operand of IL code use separate embedding models. The embedded vectors are then sequentially combined to preserve the IL structure. The proposed method was validated using Long Short-Term Memory (LSTM), LSTM-Autoencoder, and Transformer models with a dataset of normal and malicious control logic. All models achieved an anomaly detection performance with an F1 score of at least 0.81. Additionally, models adopting the proposed embedding method outperformed those using conventional embedding methods by 0.088259. The proposed control logic anomaly detection method enables the model to learn the context and structure of control logic and identify code with inherent vulnerabilities.

Anomaly detection systems for detecting cyber threats in industrial control systems (ICSs) can suffer performance degradation because rare normal samples may not be sufficiently learned. To address this problem, various data augmentation methods have been studied. However, existing methods may generate data in unknown regions based on model decision boundaries or unnecessarily augment regions that already contain sufficient samples. Many existing methods also require anomalous data. To address these limitations, we propose Tail-Guided Low-Density Normal Data Augmentation (TGLDA), which operates using only normal data. TGLDA consists of a low-density region estimator and a data generator. It identifies low-density regions with relatively few samples in the observed data distribution and selectively augments these regions. TGLDA uses feature-tail seeds as heuristic starting points for generating samples in low-density regions. Experiments on a water treatment cybersecurity dataset show that TGLDA successfully generates data focused on low-density regions compared with other data augmentation methods. When each augmentation method was applied to baseline anomaly detection models, TGLDA achieved the highest Accuracy and F1-score across all models. These results show that TGLDA can mitigate problems caused by a lack of rare normal samples in real-world ICS environments where only normal data are available.