The　finite　element　model　of　the　mega-sub　isolation　system　was　established　in　this　paper.　Incremental　dynamic　time　history　analyses　of　the　mega-sub　isolation　system　considering　different　corrosion　states　was　carried　out　by　inputting　20　actual　ground　motion　records　with　amplitude　modulation　on　the　peak　ground　acceleration　(PGA).　240　sets　of　seismic　responses　samples　were　obtained,　and　the　effect　of　steel　corrosion　on　the　seismic　performance　of　the　mega-sub　isolation　system　was　discussed.　Using　machine　learning　method　to　correlate　structural　information,　ground　motion　information,　and　structural　damage　level,　the　prediction　results　of　six　machine　learning　algorithms　on　the　damage　level　of　the　mega-sub　isolation　system　were　given.　The　overall　prediction　accuracy　of　extreme　gradient　boosting　tree,　gradient　boosting　tree,　random　forest,　and　decision　tree　reached　more　than　80%.　The　extreme　gradient　boosting　tree　algorithm　performed　the　best,　with　an　accuracy　rate　of　86.6%　and　a　higher　prediction　accuracy　for　different　damage　states.　The　prediction　accuracy　of　the　support　vector　machine　algorithm　was　the　lowest,　with　an　accuracy　rate　of　60.3%.　©　2023　Chinese　Vibration　Engineering　Society.　All　rights　reserved.