Domain names, which are indispensable for accessing Internet services, are also exploited by attackers to direct users to phishing sites or malware-infected sites. Therefore, determining whether a domain name is malicious or not is crucial for security measures. In this paper, we perform malicious domain detection using a publicly available dataset that includes statistical features of domain strings, publicly available information from sources such as WHOIS, and DNS query and response information. As a result, we achieve a recall of 0.940 and an F1-score of 0.968 with a Random Forest classifier, a recall of 0.946 and an Fl-score of 0.968 with both Random Forest and XGBoost classifiers using only statistical features, and a recall of 0.983 and an Fl-score of 0.986 with an XGBoost classifier using statistical features and DNS query and response information.
There is a technique called vulnerability chaining in which an attacker does not just target a single vulnerability but combines multiple vulnerabilities to infiltrate a target. Therefore, when evaluating product safety, it is insufficient to check only one vulnerability. A search system had been proposed to enable comprehensive retrieval of information on multiple vulnerabilities by building an ontology of information related to vulnerabilities and products. The system obtains information about vulnerabil-ities and linked products to any desired extent. In this paper, we propose a method to predict the severity of newly discovered vulnerabilities using the base CVE scores of software with related vulnerabilities through the system. Using the constructed severity prediction model, we confirmed that severity can be predicted with an accuracy of 55% to 64% for severity classification.