A Novel Distance Metric Based on Differential Evolution

Distance has been employed as a representation of similarity for half a century. Many different distance metrics have been proposed in this duration such as Euclidean, Manhattan, Minkowski and weighted Euclidean distance metrics. Each of them has its own characteristics and is calculated in different formulations/manners. In this paper, a novel distance metric, which has a high adaptation capability, was proposed. In order to increase the adaptation ability of the proposed distance metric, its parameters were optimized according to the employed dataset by differential evolution (DE), which is a metaheuristic optimization method. The proposed distance metric was employed in the k-nearest neighbor, and 30 different benchmark datasets were used in the evaluation of the proposed approach. Each of the parameters of the novel distance metric and the parameters of DE was assessed based on the obtained accuracies. Obtained results validated the applicability of the proposed distance metric and the proposed approach.