Machine learning approaches to predicting ai assistant reuse intention among university students

Abstract

The growing integration of Artificial Intelligence (AI) tools in higher education has transformed how students learn, create, and accomplish academic tasks. However, limited research has examined the behavioral and contextual factors that influence students’ satisfaction with AI assistants and their intention to reuse these tools. This study develops and evaluates predictive models to estimate student satisfaction and continued use of AI assistants using real interaction data. The dataset includes 10,000 AI-assisted learning sessions from students across multiple disciplines and academic levels. Each record captures session characteristics such as duration, number of prompts, and task type, along with AI engagement level, final outcomes, and satisfaction ratings. The target variable, reuse intention, indicates whether a student returned to use the AI assistant. Two ensemble learning algorithms, Random Forest and Gradient Boosting, were implemented to predict reuse intention and satisfaction levels. Model performance was evaluated using accuracy, F1-score, and ROC–AUC metrics. Feature importance analyses identified the most influential predictors of user engagement and satisfaction. The results reveal that the AI assistance level, session length, and discipline type are the strongest predictors of both satisfaction and reuse intention. Among the models tested, Gradient Boosting achieved the highest predictive accuracy, confirming its suitability for behavioral prediction tasks in educational contexts. These findings demonstrate the potential of ensemble learning techniques for uncovering patterns of sustained AI engagement and provide actionable insights into human–AI interaction in higher education. The study shows how behavioral data can inform the design of personalized, ethical, and effective AI-based educational tools.

Ähnliche Arbeiten