To　realize　the　safety　risk　prevention　and　control　at　the　exits　of　urban　expressway　interchange　and　improve　the　overall　safety　of　urban　road,　the　safety　risk　level　at　the　exits　of　urban　expressway　interchange　was　predicted　and　the　impacts　of　operating　conditions,　control　devices,　road　conditions,　weather,　and　other　factors　on　the　exits　of　urban　expressway　interchange　were　explored.　Based　on　navigation　data　and　field　survey　data,　the　traffic　order　index　was　used　as　the　accident　surrogate　index,　eXtreme　Gradient　Boosting　(XGBoost)　was　adopted　to　construct　the　traffic　order　prediction　model,　and　interpretable　machine　learning　framework　named　SHapley　Additive　exPlanation　(SHAP)　was　utilized　to　analyze　the　underlying　causes　of　safety　risks.　The　results　show　that　XGboost　can　accurately　predict　interchange　exits　risks,　with　an　accuracy　of　93.69%,　precision　of　93.73%,　and　recall　of　93.69%.　The　congestion　index　is　an　essential　factor　influencing　the　safety　risks　of　interchange　exits.　Easing　congestion,　reducing　traffic　diverge　and　merge　numbers,　and　weakening　the　impact　of　weather　have　a　positive　impact　on　improving　traffic　safety.　The　impact　of　the　number　of　advance　guide　signs　or　lanes　on　the　traffic　order　index　varies　at　different　congestion　states.　©　2022,　Editorial　Department　of　Journal　of　Southeast　University.　All　right　reserved.