In this work, we propose a data-driven framework to identify precursors of extreme events in turbulent reacting flows. Specifically, we tackle the problem of flashback prediction in a lean hydrogen reheat combustor. Our framework is composed of two …