Due to the difficulties of measuring concentrate and tailing grade online and unstability of the flotation performance, a data-driven optimal setting for froth image features is proposed. The froth image features are optimal set according to the type of feed ore grade. Considering the distribution nature of samples of each feed grade type, it is tried to use case-based reasoning method for obtaining the froth status with optimal flotation performance from history data. When lack of enough experiential knowledge, the interval type II fuzzy system modeling method based on the multi-centers fuzzy C-means clustering and probabilistic support vector regress method is adopted to build the concentrate and tailing grade model. Then intelligent optimization algorithm is applied to search the optimal values for the froth image features. The application to the stibium flotation process shows the effectiveness of the proposed method.