With　the　influence　of　extreme　weather,　road　surface　temperature　(RST),　which　threatens　the　safety　of　people＇s　travel,　has　attracted　more　and　more　attention　to　the　government　and　citizens.　However,　traditional　methods　are　hard　to　meet　real-time　requirements　in　forecasting　RST.　In　order　to　improve　the　predictive　accuracy　of　RST　and　meet　the　real-time　requirement,　this　paper　compares　three　different　machine　learning　algorithms　including,　Least　Absolute　Shrinkage　and　Selection　Operator　(LASSO),　Random　Forest　(RF)　and　Gradient　Boosting　Regression　Tree　(GBRT).　Using　the　RST　data　and　BJ-RUC　(Beijing-rapidly　update　cycle)　data　during　November　2012　and　June　2015,　the　performance　of　three　models　is　evaluated.　The　experimental　results　show　that　GBRT　performs　the　best　and　its　MSE　is　6.7853.