Nuclear reactor core degradation and relocation during a severe accident involve numerous coupled physical phenomena and generate extremely high-dimensional simulation outputs when analysed using integral codes such as MELCOR. Interpreting these results is often subjective and depends strongly on the analyst’s focus. In this work, we applied the Decision Tree Insights Analytics (DTIA) framework to time-series data from a severe-accident simulation of a Nordic boiling-water reactor to support a more systematic, data-driven identification of relevant variables. DTIA was used to classify accident progression into four time windows associated with key physical events. From an initial set of 23,950 MELCOR COR-package parameters, DTIA reduced the analysis to 133 variables that were most strongly associated with distinguishing these time windows. Many of the highlighted variables were consistent with established understanding of core degradation and relocation, such as mass relocation of fuel and structural materials toward the lower head. At the same time, DTIA identified patterns that are not fully explained by current interpretations, including increases in certain structural and cladding mass variables prior to core support plate failure. The results show that DTIA can systematically prioritise variables from large severe-accident datasets and help separate well-understood behaviour from observations that warrant further physical investigation. While DTIA reduces subjectivity in variable selection, interpretation of the extracted insights still requires domain expertise.