Research on Case Index BCS-Tree and Its Constructing Method

Aiming at the existing problems in case index study, a new method, called BCS-Tree, is proposed. Firstly, the GRC algorithm is improved in self-adaptive way, which can solve deficiencies of being seriously affected by initial values and just applying to convex cluster based on existing cluster methods. Then the handling ability of MBR method for the nonlinearity and no normality data is enhanced by integrating KICA with MBR. After analyzing existing method, the dual reference point selection method is designed. Moreover, the BCS-Tree constructing method is presented based on the improved GRC algorithm and dual reference point clustering splitting. Finally, based on comprehensive analyzing the possible distributing relation between query point and case data, the BCS-Tree query algorithm is designed. Furthermore, the BCS-Tree and query algorithm are analyzed by theory derivation and instance verification. The result proves that the BCS-Tree index constructing method presented in this paper is of better robustness and applicability, and BCS-Tree together with the query algorithm is of better searching efficiency. For case-based reasoning and case index research domain, the BCS-Tree provides effective method supporting and new research thoughts.