Effects of Distance Measurement Methods in K-Nearest Neighbor Algorithm to Select Indonesia Smart Card Recipient

Abstract

The Smart Indonesia Program (Program Indonesia Pintar, or PIP) through the Smart Indonesia Card (Kartu Indonesia Pintar, or KIP) is the provision of educational cash assistance to school-age children (aged 6 - 21 years) who come from poor and vulnerable families. A critical step in the KIP process is to determine the eligibility of a child who is generally selected manually. In this study, the researcher wanted to use the K-Nearest Neighbor (K-NN) algorithm to classify students eligible for KIP recipients. In addition, this paper analyzes the effect of several distance measurement functions on K-NN. There are four distance functions, that is Euclidean distance, Mahalanobis distance, Manhattan distance, and Minkowski distance. In addition, it also analyzes influence of feature selection on the performance of the classification model. The result of the K-NN algorithm has been evaluated using accuracy, precision, recall, and F-1. The study found that the combination of K-NN and Mahalanobis distance function has the highest performance compared to the other three distance functions to classify high-dimensional datasets. While for fewer variables, K-NN and Manhattan distance are the best solutions with the best performance compared to the other three distance functions. The last, feature selection can improve the performance of the classification model in any experiment scenario, except for the scenario of K-NN with Mahalanobis distance function.