Analysis of material deprivation in Poland: a machine learning approach

Abstract

The material deprivation indicators refer to the inability to afford some items considered by most people to be desirable or even necessary to lead an adequate life. Their analysis enables measuring living standards directly by means of looking at the enforced lack of necessities. Therefore, it has recently been gaining much attention in poverty research. The study aims to examine the risk of material deprivation in Poland. It is based on recent Statistics on Income and Living Conditions (EU-SILC) 2020 data. The study applies Logistic Regression and commonly used machine learning algorithms, such as Random Forests, Extra Trees Classifier, Gradient Boosting, XGBoost algorithm and AdaBoost algorithm. Due to the easy access to free software and the great advances in computational speed, machine learning methods are gaining popularity in many fields, particularly in poverty research. The analyses performed for various sets of indicators have shown that the best goodness of fit measured by AUC and accuracy has Gradient Boosting. It is found that the essential characteristics affecting the occurrence of material deprivation are income, age and level of education of household members, type of household, and presence of disabled or unemployed people.