From profiles to pathways: machine learning prediction of course preferences in Turkish adult education

In non-formal adult education systems, the lack of individualized guidance often limits the alignment between learner profiles and appropriate course pathways. This study presents a supervised learning approach to predict course preferences in Turkey’s largest municipally operated adult training initiative—İSMEK (Istanbul Metropolitan Municipality Vocational Courses)—using learner data collected between 2019 and 2023. The dataset includes demographic features (e.g., age, education, employment status, disability), enrollment records, and certification outcomes. Multiple supervised classification models, including Decision Trees (DT), Random Forest (RF), Gradient Boosting (GB), LightGBM, and ensemble methods, were employed to predict learners’ course preferences. Prediction performance was evaluated using accuracy, precision, recall, and F1-score metrics. Results indicate that learner profiles contain sufficient predictive signals to enable effective course preference modeling. Notably, including both certified and non-certified participants, as well as individuals with disabilities, improved model generalizability and fairness. The findings support the integration of predictive analytics into lifelong learning systems to enhance institutional decision-making, reduce the mismatch between learner needs and course provision, and promote equitable access. The study also operationalizes andragogical principles in data-driven educational design, offering scalable implications for policymakers and program administrators aiming to strengthen guidance in non-formal vocational education contexts.