The　expressway　is　extremely　important　to　transportation,　but　high　road-surface　temperatures　(RST)　can　cause　many　traffic　accidents.　Most　of　the　hourly　RST　prediction　models　are　based　on　numerical　methods,　but　the　parameters　are　difficult　to　determine.　Statistical　methods　cannot　achieve　the　desired　accuracy.　To　address　these　problems,　this　paper　proposes　a　machine　learning　algorithm　that　utilizes　gradient-boosting　to　assemble　a　ReLU　(rectified　linear　unit)/softplus　Extreme　Learning　Machine　(ELM).　By　using　historical　data　from　the　airport　and　Badaling　expressways　collected　between　November　2012　and　September　2014,　sigmoid　ELM,　ReLU　ELM,　soft　plus　ELM,　ReLU　gradient　ELM　boosting　(GBELM)　and　softplus　GBELM　were　applied　for　RST　forecasting,　RMSE　(root　mean　squared　error),　PCC　(Pearson　Correlation　Coefficient),　and　the　accuracy　of　these　methods　were　analyzed.　The　experimental　results　show　that　ReLU/softplus　can　improve　the　performance　of　traditional　ELM,　and　gradient　boosting　can　further　improve　its　performance.　Thus,　we　obtain　a　more　accurate　model　that　utilizes　GBELM　with　ReLU/softplus　to　forecast　RST.　For　the　airport　expressway,　our　proposed　model　achieves　an　RMSE　within　3　degrees　C,　an　accuracy　of　81.8%　and　a　PCC　of　0.954.　For　the　Badaling　expressway,　our　model　achieves　an　RMSE　within　2　degrees　C,　an　accuracy　of　87.4%　and　a　PCC　of　0.949.