Distributed multisensor networks record multiple data streams that can be used as inputs to a machine learning model designed to classify proliferation-relevant operations at nuclear reactors. This work proposes methods to assess the importance of each node (a single multisensor) and region (a group of collocated multisensors) to model accuracy. This, in turn, provides insight into model explainability, a critical requirement of data-driven applications in nuclear security. To determine the importance of the various nodes and regions for a given classification problem, traditional wrapper methods for feature importance were extended to nodes and regions in a multisensor network. On a dataset collected at the High Flux Isotope Reactor at Oak Ridge National Laboratory by a network of Merlyn multisensor platforms, these methods were used to identify high value and confounding nodes and regions for classifying nuclear reactor operational state. Specifically, the nodes near the facility’s cooling tower were identified as high value sources. When applied in conjunction with black-box classifiers such as neural networks, node and region importance can provide insight into an otherwise opaque classification model in the nuclear security domain.