An　accurate　pause　rate　prediction　is　crucial　for　calculating　the　appropriate　size　of　expressway　rest　area　This　study　introduces　an　eXtreme　Gradient　Boosting　(XGBoost)　pause　rate　prediction　model　that　utilizes　the　tree-structured　Parzen　estimator　for　Bayesian　optimization　of　its　parameters.　The　model　incorporates　13　input　variables　selected　by　considering　traffic　flow,　rest　area　location　and　functionality,　regional　development,　and　geographical　environmental　features.　These　variables　were　used　to　predict　the　hourly　pause　rate,　thereby　refining　the　temporal　granularity　of　the　prediction.　Compared　with　the　XGBoost　model　with　default　parameters,　genetic　algorithm-based　back　propagation　neural　network,　and　genetic　algorithm-based　wavelet　neural　network,　the　proposed　model　is　more　accurate,　validated　by　the　root　mean　square　error,　mean　absolute　error,　mean　absolute　percentage　error,　and　coefficient　of　determination　values.　The　confidence　intervals　of　the　proposed　model　were　estimated　using　the　bootstrap　method.　The　Shapley　Additive　explanations　model　(SHAP)　was　employed　to　rank　and　interpret　feature　importance,　thereby　enhancing　the　interpretability　of　the　model.　The　traffic　volume　and　the　relative　location　of　the　rest　areas　were　found　to　be　the　most　important　factors.　The　proposed　methodology　offers　novel　insights　and　approaches　for　designing　rest　area　during　the　planning　and　construction　phases　of　expressways.　©　2024　Elsevier　Ltd