Time Series-Based Zero-Day Vulnerability Prediction

Zero-day vulnerabilities pose significant threats to digital infrastructure, as they are exploited before developers become aware of and patch them. Proactive risk prediction can enhance cyber security by identifying vulnerabilities that are likely to be critical. This study proposes a time series-based machine learning framework for predicting the risk level of zeroday vulnerabilities using Common Vulnerabilities and Exposures. Data. We leverage natural language processing (TF-IDF) on vulnerability descriptions and train gradient-boosting models to classify vulnerabilities as high-or low-risk based on structured and unstructured features. Our approach includes SMOTE for class imbalance and trend analysis to understand the evolution of high-risk vulnerabilities.