Akademik Veri Yönetim Sistemi
Informatica (Slovenia), cilt.49, sa.27, ss.29-42, 2025 (ESCI)
Traditional vocabulary evaluation techniques frequently emphasize correctness above behavioral indications such as attempts and reaction time. To overcome this gap, our study proposes a machine learning technique that combines behavioral analysis with linguistic insights to discover vocabulary gaps among Turkish language learners. A Support Vector Machine (SVM) model was constructed with a Radial Basis Function (RBF) kernel and refined via grid search to maximize hyperparameters (C=10, γ=0.1) using a dataset of 1,000 interactions from 20 students. Behavioral attributes such as attempt count, answer response time, and answer correctness were collected to quantify student uncertainty and engagement. The approach also integrates word difficulty levels and thematic categories. An equation-based labeling technique was first applied to identify vocabulary weaknesses, laying the foundation for subsequent machine learning classification. The findings demonstrated strong performance, achieving an accuracy of 89%, precision of 86%, recall of 91%, and an F1-score of 88%, surpassing linear and polynomial kernel alternatives. These results underscore the importance of behavioral metrics in adaptive learning systems and support scalable integration into mobile applications.