Gini Index with Local Mean Based for Determining K Value in K-Nearest Neighbor Classification

Abstract

A process that explains and differentiates the data class is called Classification. The nearest neighbor is calculated based on the distance of each data, especially to determine the k value in the data. To fix K-Nearest Neighbor, it is necessary to test data class and train with Local Mean Based K-Nearest Neighbor using the closest distance measurement of Manhattan to each local mean of each data class. Gini Index is used in the process of calculating each weight in the data attribute. In this research, Gini Index, K-Fold Cross Validation and Local Mean are needed in the K-Nearest Neighbor classification. In Iris data the lowest k value is k=1, k=48, k=49, and k=50 accuracy is 94.67%, while the highest k value is k=12 and k=13 accuracy is 97.33%. So the result of the highest k value becomes the best k value in this study. Likewise with the Ionosphere data the lowest k value of k=50 accuracy is 86.92%, while the highest k value is k=2 with accuracy 92.89%, the Ionosphere data is the best k=2 and two Voice Gender and Lower Back data.