Adaptive intuitionistic fuzzy neighborhood classifier

Due to the diversity and complexity of the actual data distribution, the traditional neighborhood classifier (NEC) is weak in adapting to the global data and has low utilization of local information, which leads to the degradation of the classifier's effectiveness. To adapt NEC to the differences of different dimensions in data distribution, this paper defines attribute sensitivity and improves the purity of neighborhood information granules by weighted distance. To improve the resolution of local information, this paper constructs an intuitionistic fuzzy neighborhood classifier (IFNEC) by combining NEC with the intuitionistic fuzzy set (IFS) and defines the membership degree and non-membership degree of the object in the neighborhood to depict characteristics of local data. In IFNEC, the multi-attribute decision matrix is used in the decision-making process, which is constructed by a support function and intuitionistic fuzzy aggregation operator to filter the information with large uncertainty. Finally, taking seven data sets from UCI, and using accuracy and F1-score as evaluation indicators, we conduct a comparative experiment between NEC and IFNEC. The experimental results show that IFNEC has better performance than NEC.