ANALYSING THE DETERMINANTS OF GRADUATE UNEMPLOYMENT IN TUNISIA USING MACHINE LEARNING

Abstract

This article analyses the determinants of youth graduate unemployment in Tunisia by combining classical econometric methods (logistic regression) with three machine learning algorithms (Random Forest, XGBoost, RBF-kernel SVM) applied to an original survey of 1,200 Tunisian graduates. The econometric results reveal that female gender, belonging to the engineering field, and education employment mismatch are the most significant determinants. The machine learning analysis confirms the predominance of gender in discriminating between unemployed and employed individuals, and uncovers non-linear relationships that parametric models fail to capture. XGBoost and SVM offer the best predictive performance. These findings call for a deep reform of the university system, targeted policies against gender discrimination, and improved recruitment transparency.