Incorporating Association Patterns into Manifold Clustering for Enabling Predictive Analytics

The goal of this research is to develop a predictive analytics technique based on manifold clustering of mixed data type. In this research, we explore the concept of statistically significant association patterns to induce an initial partition on data for deriving manifolds. Manifolds are hyperplanes embedded in low dimensions. The advantage of this novel technique is a bootstrap on data clusters that reveals statistical associations from the information-theoretic perspective. As an illustration, the proposed technique is applied to a real data set of diabetes patients. An assessment on the proposed technique is performed to investigate the effect of bootstrap based on association patterns. Results of the preliminary study demonstrate the feasibility of applying the proposed technique to real-world data.