Let the data speak about the cut-off values for multidimensional index: Classification of human development index with machine learning

Abstract

The Human Development Index (HDI) classification is essential as it relates to international aid policies and business strategies. Although the existing literature has criticized the arbitrariness of cut-off values of the HDI, few proposed an ideal approach to overcome this drawback. This paper first employs the unsupervised machine learning techniques, the K-means clustering and Partitioning Around Medoids algorithms, to cluster the HDI and offers more reasonable cut-off values for classifying countries in combination with the current HDI calculation method. The results indicate that we can group the countries worldwide into three clusters, given the 2018 HDI dataset. We suggest cut-off values of 0.65 and 0.85 to classify low, medium, and high human development countries. This paper provides a new perspective to classifying the HDI based on the similarity of countries’ development but not subjective judgments.