Cost sensitive multi-class fuzzy decision-theoretic rough set based fault diagnosis

Cost-sensitive multi-class fuzzy decision-theoretic rough set (MC-FDTRS) is proposed by considering cost sensitivity in practical fault diagnosis process. MC-FDTRS generalizes indiscernibility relation of multi-class decision-theoretic rough set to Gaussian kernel based fuzzy equivalence relation, and then it can deal with numerical data directly without discretization. MC-FDTRS introduces cost matrix for loss functions to solve classification problems where different types of misclassification have different costs. Positive, negative and boundary region classification rules are extracted from data based on Bayesian risk minimum principle. MC-FDTRS method is applied to fault diagnosis of power transformer to validate the effectiveness of the proposed method. Experimental results on transformer fault diagnosis examples show that the proposed method is well suitable for cost-sensitive fault diagnosis tasks and leads to lower misdiagnosis cost.