The exact predication of rockburst is a foundation for the disaster harness of rockburst and engineering measures. An improved calculation model for the predication of rockburst using the theory of set pair analysis (SPA) based on data mining is discussed. A model for data mining in fuzzy difference coefficient using the concept of the approach degree is investigated during the procedure of SPA. The results from the application to practical example and comparisons with other methods are fairly good. And they also show that the improved method of SPA used to forecast intensity of rockburst is feasible and effective.