Comparing k-nearest neighbor and k-means methods for clustering Indonesian farmers’ welfare

Abstract

A farmer’s welfare classification can be performed to accommodate all significant issues that will assist policymakers, government, and scientists. This study aims to compare K-Nearest Neighbor (K-NN) and K-Means methods for clustering Indonesian farmers’ welfare using the fifth wave of Indonesia Family Life Survey (IFLS 5) data. The K-Means method is an unsupervised learning algorithm by classifying the data according to the closest distance between observed and centroids. The K-NN method is a supervised learning algorithm by classifying most of the nearest neighbour data. This study used fifteen factors affecting farmers’ welfare including land area, type of water, type of rice, income, expenditure, loan, mobile phone use, harvest frequency, crop failure, land ownership, gender, age, level of education, home ownership, and ownership of health insurance. The K-NN performed well to classify farmers’ welfare as the K-Means methods in the district data, with an accuracy of 89.8% compared to 53.7%. The K-NN classification results in provinces data showed that the provinces of Bali, East Java, South Kalimantan, Lampung, West Nusa Tenggara, South Sulawesi, and South Sumatra were included as prosperous provinces; while the provinces of Banten, DI Yogyakarta, West Java, Central Java, West Sumatra, and North Sumatra were included as non-prosperous provinces.