High　temperatures　can　deteriorate　the　mechanical　properties　of　geopolymer　buildings.　Rapid　prediction　of　the　residual　mechanical　properties　and　failure　mode　of　geopolymer　concrete　(GPC)　after　thermal　treatment　is　important　to　ensure　the　safety　of　buildings.　In　this　study,　a　method　combining　BP　neural　network　with　numerical　simulation　is　proposed　to　predict　the　residual　mechanical　properties　and　failure　mode　of　GPC　subjected　to　high　temperature　simultaneously.　15　GPC　specimens　with　65　%　aggregate　content　were　prepared　by　mixing　metakaolin-based　geopolymer　and　limestone　aggregates.　These　GPC　specimens　were　divided　into　5　groups　and　heated　at　25　°C,　200　°C,　400　°C,　600　°C　and　800　°C,　respectively.　The　uniaxial　compression　tests　were　conducted　on　the　GPC　after　thermal　treatment.　The　variation　in　the　peak　stress,　peak　strain　and　elastic　modulus　of　GPC　with　temperature　was　further　analyzed.　The　prediction　method　by　combining　the　BP　neural　network　with　numerical　simulation　was　constructed　to　predict　the　mechanical　properties　and　failure　mode　of　GPC　subjected　to　high　temperature.　The　reliability　of　the　prediction　method　was　verified　by　comparing　the　predicted　results　and　the　experimental　results.　The　results　show　that　as　the　temperature　increases,　the　peak　stress　and　elastic　modulus　decrease　and　the　peak　strain　increases,　showing　a　deterioration　of　the　mechanical　properties　of　GPC　after　thermal　treatment.　The　prediction　method　combining　the　artificial　neural　networks　and　numerical　simulation　has　high　accuracy,　with　prediction　errors　of　less　than　10　%　for　peak　stress　and　less　than　20　%　for　peak　strain　and　elastic　modulus.　The　results　also　show　that　under　uniaxial　compression　load,　the　predicted　failure　mode　of　GPC　after　thermal　treatment　is　consistent　with　the　experiment　results.　The　method　proposed　in　this　study　can　simultaneously　predict　the　mechanical　properties　and　failure　mode　of　GPC　with　low　error,　which　have　potential　application　prospects　in　building　safety　assessment.　©　2023　Elsevier　Ltd