Abstract
Cyber-attacks have increased in number and severity, which has negatively affected businesses and their services. As such, cyber security is no longer considered merely a technological problem, but must also be considered as critical to the economy and society. Existing solutions struggle to find indicators of unexpected risks, which limits their ability to make accurate risk assessments. This study presents a risk assessment method based on Machine Learning, an approach used to assess and predict companies' exposure to cybersecurity risks. For this purpose, four algorithm implementations from Machine Learning (Light Gradient Boosting, AdaBoost, CatBoost, Multi-Layer Perceptron) were implemented, trained, and evaluated using generative datasets representing the characteristics of different volumes of data (for example, number of employees, business sector, and known vulnerabilities and externel advisor). The quantitative evaluation conducted on this study shows the high accuracy of Machine Learning models and Especially Multi-Layer Perceptron was the best accuracy when working compared to previous work.
