An Interactive Clustering Methodology for High Dimensional Data Mining

This study develops an interactive clustering model and methodology for high dimensional data. The similarity index is calculated with proposed formulation for both continuous-scaled and nominal-scaled attributes. The associated similarity score values are constructed into a graph as clique partitioning problem, which can be reformulated into a form of unconstrained quadratic program model and then solved by a Tabu search heuristic incorporating strategic oscillation with a critical event memory. The complexities of high dimensional data mining are discussed from both mathematical modeling and computational algorithm points of view.