This paper proposes the modified time series data mining framework applying Reconstructed Phase Space to construct clusters from the temporal patterns, which are predictive of interesting events. Cluster objective function used in the presented technique is defined not only by cluster internal predictive patterns but also by estimation of the efficacy of cluster to characterize the predictive clusters. For prediction stage, framework uses initial both information about predictive clusters and expert knowledges by applying Sugeno-type fuzzy inference. Experimental results demonstrate presented framework can reach more effective results than existed algorithms, which utilize reconstructed phase space.