Many machine learning methods have been successfully applied in fault diagnosis of industrial processes, while the integrated feature space has not been fully considered and utilized. In this paper, a fault diagnosis method with integrated feature space and optimized random forest is proposed to realize accurate diagnosis. By means of the integration of static and dynamic information, an embedded & wrapped feature selection process is designed to provide the optimal feature combination for classifier. In addition, the optimal feature combination and hyper-parameter optimization are utilized to construct optimized random forest to improve the generalization of the model. Experimental results on Tennessee Eastman benchmark show that the proposed method outperforms traditional approaches with accuracy and F1 score that exceed 87% and 88% respectively.