Applying Machine Learning to Longitudinal Administrative Data: A Case Study in Education

This paper presents the results of an extensive empirical evaluation study of several machine learning methods used to predict the attained level of education (ALE) in Italy based exclusively on longitudinal administrative data. The assessment compares model predictions with observed ALE values available in administrative sources, evaluating both distributional preservation and unit-level prediction accuracy. Our findings show that machine learning techniques can effectively exploit longitudinal administrative information to generate accurate and coherent ALE estimates, providing valuable support for the production of register-based statistics.