Generating human-centered explanations is essential for creating applications usable by non-ML expert users. In this paper, we incorporate human-derived knowledge into a model predicting the severity of COVID-19 spread to generate explanations using rules that align with users’ intuition and logic. Using LORE, a post-hoc, agnostic explanation methodology, we developed a specialized algorithm to generate a synthetic neighborhood that closely resembles the training data. We validate this algorithm’s quality by comparing its results with neighborhoods produced by the in-built generator of LORE. The custom neighborhood generator is then used to train a surrogate model, from which general explanations are derived as logical predicates. Finally, we propose a visualization mock-up for the generated rules.