The　issue　of　stroke　happening　in　world　range　acting　a　significant　problem　for　human　health.　Until　2019,　there　were　about　411　death　cases　per　day　in　America　which　is　3　years　earlier　than　the　appearance　of　e-cigarettes　in　2022.　However,　the　judgement　between　different　doctors　could　be　variable　and　a　slightly　human　error　in　complex　situations　could　lead　to　a　bad　estimation　result.　To　deal　with　this,　the　prediction　of　stroke　probability　of　patients　could　provide　an　objective　reference　for　doctors.　This　study　made　a　data　treatment　of　factors　choosing　and　translation　then　calculating　4　accuracy　of　models　and　finally　an　application.　To　be　more　specific,　an　online　stroke　dataset　was　employed　for　factor　analysis　firstly　and　three　machine　learning　algorithms　were　utilized　to　select　most　important　stroke-related　factors.　6　most　influential　factors　were　further　passed　into　four　machine　learning　models　for　training.　The　work　gave　the　result　of　the　highest　accuracy　performance　of　Random　Forest　between　the　four　models　which　could　be　assistance　of　the　prediction　in　stroke.　©　2022　SPIE.