In　light　of　the　increasing　amount　of　traffic　disorder　at　road　traffic　hubs,　to　improve　traffic　safety,　it　is　essential　to　detect　road　risks　in　advance　and　analyze　the　causes　after　its　occurrence.　In　this　study,　interpretable　machine　learning　(ML)　is　employed　to　analyze　the　traffic　order　by　using　a　set　of　multi-source　data　comprising　traffic　conditions,　traffic　control　devices,　road　conditions,　and　external　conditions.　Data　were　collected　from　the　exits　of　some　Beijing　expressways　via　navigation　and　field　investigations.　The　traffic　order　index　(TOI)　based　on　aggregate　driving　behavior　data　is　used　as　a　new　surrogate　index　to　evaluate　the　safety　risk.　A　traffic　order　prediction　model　is　then　constructed　by　adapting　the　eXtreme　Gradient　Boosting　(XGBoost)　ML　method.　In　addition,　SHapley　Additive　exPlanation　(SHAP)　is　employed　to　interpret　the　results　and　explore　the　relationships　between　the　influencing　factors　and　the　traffic　order.　The　results　indicate　that　XGBoost　could　predict　the　traffic　order　levels　well,　and　achieved　an　accuracy,　precision,　recall,　and　F-1-score　of　92.62%,　92.67%,　92.62%,　and　92.63%,　respectively.　The　congestion　index　was　found　to　have　a　great　influence　on　traffic　order.　Furthermore,　the　number　of　lanes,　advance　guide　signs,　and　weather　conditions　can　have　different　effects　on　the　traffic　order　under　different　traffic　conditions.　(C)　2022　Elsevier　B.V.　All　rights　reserved.