This　paper　aims　to　optimize　the　intake　characteristics　of　a　side　ported　Wankel　rotary　engine　by　combining　machine　learning　(ML)　with　genetic　algorithm　(GA).　The　computational　samples　are　generated　using　Sobol　se-quences,　in　which　the　variables　are　the　timing　of　port　full　opening,　port　start　closing,　and　port　full　closing　(PFC).　A　two-layer　structured　ML　prediction　model　is　establishedwith　the　intake　phases　and　geometric　parameters　as　input　variables.　The　results　show　that　the　coefficients　of　determination　of　the　prediction　models　built　by　Gaussian　process　regression　are　greater　than　0.99.　The　response　surface　presents　that　the　PFC　timing　determines　the　intake　loss　and　volumetric　efficiency　compared　to　others.　The　volume　efficiency　and　intake　loss　are　fitted　as　a　quadratic　function　in　the　Pareto　front.　In　all　the　typical　cases,　the　deviation　between　prediction　and　calculation　is　less　than　1%.　In　the　typical　case　C,　the　intake　loss　is　reduced　by　19.39%,　and　the　volumetric　efficiency　is　only　reduced　by　0.01%.　It　is　promising　to　integrate　ML　with　GA　for　further　improvements　of　engine　performance.