As　the　increasing　requirement　of　compact　and　light　electronic　and　electrical　system,　especially　for　high　energy　density　storage　system,　the　development　of　high　dielectric　permittivity　materials　becomes　one　of　critical　solutions.　Among　all　materials,　barium　titanate　materials　have　been　found　to　be　a　promising　ceramic　capacitor　dielectric　system,　like　BaTi1-xHfxO3　material　and　BaTi1-xSnxO3　material.　However,　to　search　the　optimal　composition,　mass　samples　are　needed　by　the　traditional　method　of　exhaustion,　which　increases　the　cost　of　work.　In　order　to　access　the　optimal　composition　of　high　permittivity　efficiently,　a　machine　learning　prediction　is　employed　to　the　searching　process　in　Sn/Ca　doped　barium　titanate　ceramics.　The　machine　learning　prediction　is　iteration　between　theoretical　predictions　and　experiment　data.　According　to　our　prediction,　the　searching　result　shows　Ba0.86Ca0.14Ti0.86Sn0.14O3　has　the　peak　values　of　dielectric　permittivity　with　ϵr=2.0×104　at　T=13°C.　As　a　result,　sample　preparation　number　reduces　to　5.5%　compared　with　the　traditional　method　of　exhaustion.　In　addition,　good　temperature　stability　has　been　found　in　Ba0.86Ca0.14Ti0.86Sn0.14O3　because　of　the　relaxor　behavior.　©　2018　IEEE.