PREDICTION OF TRANSITION PROBABILITIES FROM UNEMPLOYMENT TO EMPLOYMENT FOR TURKEY VIA MACHINE LEARNING AND ECONOMETRICS: A COMPARATIVE STUDY

Abstract

In this study, it is mainly aimed to predict transition probabilities of individuals who are previouslyunemployed and get employment or stay unemployed. In order to do that, Household Labor ForceSurveys conducted in Turkey are merged and matched from 2004 to 2016. Information aboutindividuals only consists of individual characteristics and qualifications since there should not be anyinformative clue about the present situation. To predict those probabilities, logistic regression analysisas econometric approach, a shallow neural network and machine learning classification algorithms arerun in order to compare them. The results indicate that classification in machine learning is slightlybetter than logistic regression and shallow neural network. While XGBoost classifier and RandomForest get 67% accuracy, logistic regression can predict only 63% of an individual’s transition andshallow neural network forecasts 51%.