Abstract
Extreme complex events and the corresponding abnormal statistics of cyber security are ubiquitously observed in many real-time systems, and the development of efficient tools to explain and properly anticipate such representative features remains a great issue. Art and science must be carefully balanced in order to determine the risk of cyberattacks. Once the risk variables have been identified, a risk assessor typically begins by gathering relevant information for each. Logs, architecture diagrams, network topology, compliance evaluations, incidents, vulnerability evaluations, threat modelling, and control assessment are all sources of information for the assessor. The assessor uses approved impact and likelihood tables to evaluate risk factors based on evidence gathered and a methodology that has been approved. Assessments that are accurate are those in which draw conclusions from large amounts of acquired data and then apply those conclusions to the calculated risk severity. Contextualizing risk requires the assessor to draw on past experience, knowledge, and observations of the system itself. Qualitative assessments of cybersecurity risks are performed. It is difficult, if not impossible, to obtain meaningful quantitative measures of cybersecurity risk variables. Risk assessments in cybersecurity cannot use quantitative estimations because they're time- and labour-intensive. A shortage of competent individuals, the quantity of staff necessary, the assessment time, and business objectives all limit the ability to scale quantitative and qualitative risk assessments. Using machine learning (ML) to forecast the severity of future risks based on previous risk assessments may provide a solution to the scalability problem in risk assessment. The intuition, insight, and skill that risk assessors use to determine the severity of a risk are all included into machine learning algorithms. As an initial step, machine learning can be used to assess risk, and if the level of risk exceeds a predetermined threshold, additional steps can be taken. The algorithm learns from each manual analysis, reducing the need for human interaction dramatically.