Breakthrough　research　is　groundbreaking　and　transformative　scientific　research　that　can　lead　to　new　frontiers　and　even　trigger　substantial　changes　in　the　scientific　paradigm.　Early　identification　of　breakthrough　research　is　crucial　for　scientists,　R&D　experts,　and　policymakers.　＇Sleeping　Beauty　in　Science＇　is　a　category　of　papers　characterized　as　＇delayed　recognition＇,　which　is　considered　as　the　crucial　carriers　of　breakthrough　research.　Machine　learning　methods　can　extract　and　learn　high-quality　information　from　a　priori　knowledge　to　predict　future　trends.　In　this　paper,　to　address　the　shortcomings　of　existing　studies　on　the　early　identification　of　breakthrough　research,　we　propose　a　framework　for　identifying　breakthrough　research　from　sleeping　beauties　using　machine　learning.　In　this　framework,　we　first　construct　machine　learning　models　to　obtain　the　relationship　patterns　between　historical　sleeping　beauties　and　their　citation　trends.　Then,　we　use　these　relational　patterns　to　identify　potential　sleeping　beauties.　Secondly,　we　construct　a　breakthrough　index　based　on　the　essential　features　of　breakthrough　research,　then　we　apply　it　to　identify　breakthrough　research　among　potential　sleeping　beauties,　enabling　the　early　identification　of　breakthrough　research.　Finally,　an　empirical　study　is　conducted　in　the　chemistry　research　field　to　verify　the　validity　and　flexibility　of　this　framework.　The　results　show　that　the　framework　can　effectively　identify　breakthrough　research　from　sleeping　beauties.　This　paper　contributes　to　the　early　identification　of　breakthrough　research,　evaluating　academic　results,　and　exploring　research　frontiers.　Additionally,　it　will　also　provide　methodological　support　for　the　decision-making　of　R&D　experts　and　policymakers.　©　2024