Methods, apparatus, and program products that can predict misconfigurations in a computing system using machine learning are disclosed herein. One method includes labeling one or more graph nodes or link nodes of a data graph of a computing system that includes one or more security vulnerabilities with a node label or link label, respectively, in which each node label represents the first security vulnerabilities associated with a particular graph node and each link label represents the second security vulnerabilities associated with a particular link node. The method further includes utilizing the graph node(s) or the link node(s) to train a machine learning algorithm to predict one or more misconfigurations in the computing system based on the security vulnerabilities and determining one or more modifications to the computing system for mitigating the one or more misconfigurations. Apparatus and program products that include and/or perform the methods are also disclosed herein.