Evidential K-NN classification with enhanced performance via optimizing a class of parametric conjunctive t-rules

Dempster's rule of combination is commonly used to pool distinct/independent bodies of evidence in the evidential k-nearest neighbor (K-NN) classifier, which sometimes limits the performance of this classifier. To solve this problem, we propose a class of parametric conjunctive combination rules based on a new family of triangular norms with selectable functions and tunable parameters. We show that the performance of the evidential K-NN classifier can be enhanced via this class of so-called parametric conjunctive t-rules when appropriate functions and parameters are selected. Numerical simulations validate our conclusions.