The　purpose　of　this　paper　is　a　machine　learning　model　that　could　predict　the　strabismus　surgery　parameter　through　the　data　of　patients　as　accurately　as　possible.　A　strabismus　surgery　parameter　design　model＇s　input　is　a　Medical　records　and　return　is　a　surgical　value.　The　Machine　learning　algorithms　is　difficult　to　get　a　desired　result　in　this　process　because　of　the　small　amount　and　uneven　distribution　strabismus　surgery　data.　This　paper　enhanced　the　data　set　through　a　WGAN-GP　model　to　improve　the　performance　of　the　LightGBM　algorithm.　The　performance　of　model　is　increased　from　69.32%　to　84.52%.　©　Published　under　licence　by　IOP　Publishing　Ltd.