Recently,　microgrids　(MGs)　have　been　attracted　more　attention　due　to　their　technique　and　economic　advantages.　However,　along　with　these　advantages,　because　of　the　cyber　and　physical　structure　of　MGs,　they　are　more　prone　to　cyber　and　physical　attacks.　To　this　end,　in　this　paper,　a　new　machine　learning　framework　is　developed　to　detect　and　mitigate　the　fake　data.　More　specifically,　a　new　machine　learning　technique　has　been　developed,　which　is　mainly　based　on　the　long-short　term　memory　(LSTM);　however,　modified　with　recurrent　neural　network　(RNN)　and　prediction　intervals　(PIs).　Finally,　an　evolutionary　algorithm　has　been　used　to　address　the　nonlinearity　and　complexity　associated　with　the　problem.　The　proposed　framework　is　tested　on　real　MG　data.　Results　show　the　efficiency　and　merit　of　the　proposed　techniques,　compare　to　the　conventional　techniques.　©　2021　Elsevier　Ltd