A Novel Classification Approach Based on Integrated Connection Cloud Model and Game Theory

Abstract The classification of problems under uncertainty environments involves the precise description and weighting of fuzzy and random indicators. Although the one-dimensional normal cloud model can effectively handle the fuzziness and randomness of the indicator in an infinite interval, it may ignore the changing tendency of the rating boundary, and become complex with the increment of indicator and sample numbers. To overcome above shortcomings, a novel classification based on the integrated connection cloud model and game theory was proposed here to depict fuzzy indicators randomly distributing in finite intervals and simultaneously embody the importance and inherent information of indicators in harmony. Namely, the connection numbers theory and the game theory based combination weighting method were first utilized to aggregate the connection cloud model of the individual indicator. Next, integrated connection clouds were simulated to express the interval-valued classification standards, and the connection degrees were calculated to identify the rank from the identity, discrepancy, and contrary relationships. Finally, the validity and feasibility of the approach proposed here were further verified by example application to the classification of surrounding rock stability and comparisons with other methods. Results indicate that the integrated cloud model based classification approach presents a precise description of fuzzy indicators distributing randomly in the asymmetrical intervals and better computational efficiency relative to the classification method using the one-dimensional cloud model. It also minimizes errors caused by the multiplication singularity point, human subjectivity and neglect of the intrinsic information of indicators.