Vulnerability Risk Prediction for Unexploited Network Weaknesses

The patent inverts the conventional approach by training the machine learning model on exploited vulnerabilities (one-class data) to predict unexploited vulnerabilities. Instead of requiring balanced datasets of both exploited and unexploited vulnerabilities, the model learns the characteristics of exploited vulnerabilities and uses this knowledge to identify similar patterns in unexploited vulnerabilities, thereby solving the one-class data limitation.

The patent introduces an intermediary mechanism where the machine learning model acts as a mediator between the one-class exploited vulnerability data and the prediction task for unexploited vulnerabilities. The model transforms the one-class data into predictive capabilities by learning underlying patterns and applying them to assess exploitation likelihood across both exploited and unexploited vulnerability classes.

The patent applies local quality by providing different priority levels and response strategies for different vulnerabilities based on their individual exploitation likelihood scores. Instead of uniform patching, each vulnerability receives targeted attention proportional to its assessed risk, allowing resources to be concentrated on high-risk vulnerabilities while lower-risk ones receive deferred or reduced attention.

The patent changes the parameter of vulnerability prioritization from binary (patched/not patched) to a continuous spectrum based on exploitation likelihood probability. This parameter transformation enables nuanced resource allocation where patching decisions are guided by predicted exploitation probabilities, allowing organizations to optimize security investments based on quantitative risk assessment.