Improving the precision of KNN classifier using nonlinear weighting method based on the spline interpolation

Precision improvement of the classifiers is one of the main challenges for the Artificial Intelligence researchers. Feature weighting is one of the most common ideas in this area. In this study, in order to increase the accuracy of the K-Nearest Neighbors (KNN) classifier, a nonlinear feature weighting method based on the Spline interpolation is used. In this approach, a unique nonlinear function is estimated for each feature. In order to find the best estimated parameters of the nonlinear function which is suitable for each feature, the evolutionary Genetic Algorithm is applied. Numerical results show that the nonlinear weighting method increases the accuracy of the classifiers compared to the linear weighting method.

I. Introduction

The definition of classifying is categorizing something or someone into a certain group or system based on certain characteristics. Each group represents the relationship between the studied subject and the knowledge objects. Nowadays, the increase of intelligent systems usage in the technology, make the classification popular. Some classification application can be referred to [1], [2]. Different classifiers have been presented, such as Neural Networks, decision tree, Naive Bayes classifier, and KNN classifier [3]. Data preparation is one of the most important issues in the classification. Data preparation includes some methods to improve the classifier efficiency, such as instance and feature selection [4]–[9]. Another powerful method is feature weighting [10].