Blind quantitative steganalysis is about revealing more details about hidden information without any prior knowledge of steganograghy. Machine learning can be used to estimate some properties of hidden message for blind quantitative steganalysis. We propose a quantitative steganalysis method based on fusion of different steganalysis features and the estimator relies on gradient boosting. Experimental result shows that our proposed method has good performance for quantitative steganalysis.