The ultimate recovery factor is of great importance to oilfields exploration,development and production.However,it is affected by numerous geological,fluidic and engineering factors and is therefore difficult to predict and find its controlling factors.The objective of this study is to forecast the ultimate recovery factor and to find out the controlling factors that influence the recovery factor of 867 onshore oil reservoirs in USA by four machine learning techniques such as gradient boosting,multiple linear regression,SVM and deep learning techniques.