Explainable AI-based Classification of Students’ Trust and Behavioral Intentions Toward Generative AI in Core Academic Subjects
Year: 2026
Type: conference-paper
Study design: Observational Study
Abstract: Generative artificial intelligence (GenAI) is increasingly integrated into academic learning, yet students differ in the level of trust they place in these systems and in their willingness to continue using them. This study developed an explainable machine-learning framework for classifying students’ trust and behavioral intention toward GenAI in core academic subjects. Survey data were obtained from 744 students in Northern Luzon, Philippines, and related questionnaire responses were transformed into meaningful features representing academic usefulness, ease of use, perceived reliability, perceived risk, ethical awareness, AI literacy, GenAI usage intensity, and subject utilization. Three gradient-boosting algorithms, namely XGBoost, LightGBM, and CatBoost, were evaluated for classifying students into Low, Moderate, and High levels of trust and behavioral intention. Results showed that CatBoost achieved the strongest performance, with 84.56% accuracy and 84.72% macro F1-score for trust classification, and 85.91% accuracy and 86.05% macro F1-score for behavioral intention classification. SHAP analysis further showed that perceived reliability, academic usefulness, AI literacy, and perceived risk were among the strongest contributors to trust, while academic usefulness, GenAI usage intensity, ease of use, and AI literacy were more influential in behavioral intention. The findings demonstrate that explainable gradient-boosting models can provide both accurate classification and interpretable insights into students’ perceptions of GenAI. The proposed framework may support educators and institutions in developing responsible, evidence-based strategies for GenAI integration in higher education.
Language: English
Subfield: Education
Field: Social Sciences
Domain: Social Sciences
SDG: Quality education
Open Access status: hybrid
OpenAlex ID: W7224938987