With　the　development　of　cell　phone　technology,　smartphone　nowadays　is　no　longer　limited　to　communication　but　is　also　well　used　in　fields　such　as　traffic　engineering,　where　they　can　be　used　to　collect　driving　vibration　data　to　evaluate　pavement　performance.　The　approach　based　on　smartphones　could　effectively　reduce　the　cost　of　pavement　detection,　in　which　the　relationship　between　smartphone　test　data　and　pavement　performance　is　the　essence　of　the　method.　This　study　predicted　the　international　roughness　index　(IRI)　based　on　an　improved　PSO-BP　neural　network　method　using　the　driving　vibration　data　collected　by　a　self-developed　smartphone　application.　The　results　show　that　the　driving　speed,　acquisition　frequency,　experimental　vehicle,　and　the　attitude　of　mobile　phones　have　a　significant　impact　on　the　vibration　data.　The　regression　value　(R)　of　the　roughness　optimized　by　PSO　is　0.90,　and　the　average　relative　error　is　18.01%,　which　is　improved　by　4.29%　compared　with　the　traditional　BP　neural　network.　The　pavement　roughness　can　be　more　accurately　predicted　by　using　the　method　proposed　in　this　study,　which　provides　theoretical　support　for　the　further　development　of　the　intelligent　detection　of　pavement　performance.　　©　2023　ASCE.