Vulnerability Time Series Forecasting: A Comparative Study of Hierarchical and Non-Hierarchical Approaches

Cybersecurity has become an increasing priority in large organizations due to the rapid evolution of digital threats. With increasingly sophisticated attacks, vulnerability management emerges as one of the main strategies to protect systems and data. However, the complexity and dynamics of threats make management processes challenging, requiring efficient and adaptable approaches. Artificial Intelligence (AI) and Machine Learning (ML) are promising tools to anticipate and mitigate these threats, offering predictive capabilities based on historical data. This article proposes the use of time series in a hierarchical manner for the forecasting of vulnerabilities in systems. The proposed methodology aims to deal with the complexity of the data, allowing a hierarchical structure to understand and capture the interdependencies and patterns among different levels more completely. The evaluation is carried out with different ML models, such as LSTM, RNN, MLP, among others, comparing the performance of hierarchical and non-hierarchical approaches. The results indicate that, in a hierarchical structure, especially LSTM, shows superior performance in vulnerability forecasting. However, in some scenarios, when using reconcilers, models like N-BEATS also demonstrate good results.