Trip　purpose　is　vital　to　infer　travel　behavior　and　predict　travel　demand　for　transportation　planning.　Therefore,　trip　purpose　prediction　has　been　becoming　an　important　field　of　research　that　can　improve　people＇s　travel　efficiency　through　travel　information,　such　as　travel　mode,　time,　location　and　so　on.　However,　there　are　a　few　challenges　linked　with　collecting　data　via　the　surveys　and　the　spatial　complexity　of　human　travel.　To　solve　the　above　problems　effectively,　the　study　adopts　GPS　data　and　land　use　data　and　proposes　a　machine　learning　method　and　prediction　model　as　forecasting　process.　The　prediction　model　is　used　to　automatically　predict　trip　purpose,　while　the　machine　learning　method　is　used　to　constantly　updating　the　prediction　model,　based　on　surveys　from　participants.　Compared　with　traditional　models,　the　method　can　significantly　improve　destination　prediction　accuracy　by　dynamically　updating.　In　addition,　the　estimation　model　is　developed　employing　the　Markov　model,　the　structure　of　model　can　use　for　a　short　training　period.　Meanwhile,　the　research　can　apply　to　crowded　place　analysis　or　in　trip　distribution　prediction,　which　shows　a　broad　application　in　transportation　planning　and　management.