Self-similarity based lightweight intrusion detection method

There are many security concerns such as data leakage, unauthorized access from outside the system and abnormal activities from inside the system. To detect these system's abnormal activities or misuse by malicious attackers, intrusion detection system (IDS) is usually adopted. Even though detection algorithms and their performance are improved, IDS still consume system resources not ignorable. For providing high performance computing environment, lightweight anomaly detection method is needed today. In this paper, we propose self-similarity measures for lightweight IDS. For normal systems, a regular and periodic self-similarity can be observed in a system's internal activities such as system calls and process status. On the other hand, outliers occur when an anomalous attack happens, and then the system's self-similarity cannot be maintained. Therefore monitoring the changes of a system's self-similarity can be used to detect the system's anomalies. From this viewpoint, we developed a new measure based on cosine similarity and found the optimal time interval for estimating the self-similarity of a given system. As a result, we can detect abnormal activities using only a few resources.