Dynamic Data Updates and Weight Optimization for Predicting Vulnerability Exploitability

With the purpose of managing efficiency in a large number of published vulnerabilities, the time-intensive process requires significant effort and efficient vulnerability prioritization procedures. In order to enhance the cybersecurity defense posture, it becomes imperative to ensure alignment with risk tolerance and business needs, making vulnerability management critical for reducing cybersecurity risks, avoiding system breaches, and optimizing organizational resources. However, traditional methods often struggle with accuracy due to static datasets and inconsistent scoring methodologies, leading to challenges in prioritizing vulnerabilities and exposing systems to significant risks. This study aims to propose a novel approach for predicting the exploitability of vulnerabilities using dynamic, continuously updated datasets and advanced learning models. A novel exploitability scoring equation is introduced and rigorously assessed through four methods: the Analytic Hierarchy Process, regression analysis, and two supervised learning models. The methodology incorporates machine learning and deep learning models to compute and predict exploitability scores, following the initial data processing and scoring calculation steps. The research framework integrates data ingestion, cleaning, and correlation processes to ensure data accuracy. It employs a dynamic dataset that is continuously updated with enriched data from diverse sources such as the National Vulnerability Database, ExploitDB, MITRE ATT&CK, and Vulners. Experimental results demonstrate a significant improvement in model training, with an accuracy of 84%. Furthermore, the predictive model achieved 82.9% accuracy in classifying vulnerabilities as exploitable or not, highlighting the importance of dynamic datasets and the comprehensive framework used to enhance vulnerability.