Anticipating Delays in Recruitment: Explainable Machine Learning for the Prediction of Hard-to-Fill Online Job Vacancies

Abstract

Online job vacancy (OJV) platforms have transformed the labor market by enabling employers to advertise jobs to a wide audience. Particularly in tight labor markets, quickly identifying vacancies likely to suffer prolonged durations is crucial. This study utilizes data from the Flemish public employment service's OJV platform to examine the effectiveness of machine learning in predicting hard-to-fill vacancies. We achieve notable predictive performance with XGBoost in forecasting recruitment delays and demonstrate the importance of capturing non-linear patterns in OJV data. SHAP (SHapley Additive exPlanations) values reveal that the textual content of vacancies and latent company characteristics are key predictors of hiring delays. Counterfactual-SHAP insights provide practical guidance for refining recruitment strategies, enhancing labor market forecasts, and informing targeted policies.