Measuring senior high school students’ use of GenAI using structural equation modeling and artificial neural networks

Abstract Generative artificial intelligence (GenAI) is increasingly being used in education. However, empirical evidence on senior high school students’ adoption of specific GenAI tools remains limited. This study examines senior high school students’ behavioral intention and self-reported use behavior toward DeepSeek, the focal GenAI tool, using the Unified Theory of Acceptance and Use of Technology 2 (UTAUT2) framework. Data were collected from 882 Indonesian senior high school students through an online questionnaire. Partial least squares structural equation modeling (PLS-SEM), artificial neural network (ANN) analysis, and importance–performance map analysis (IPMA) were used to test the hypothesized relationships, examine predictive importance, and identify practical priorities. The results revealed that performance expectancy, effort expectancy, hedonic motivation, and facilitating conditions significantly predicted students’ behavioral intention, whereas habit and social influence were not significant predictors. Both behavioral intention and facilitating conditions also influenced students’ self-reported use behavior. The ANN analysis confirmed the strong predictive power of facilitating conditions and performance expectancy, with IPMA highlighting performance expectancy as the highest-performing factor. These findings indicate that students’ adoption of DeepSeek is driven more by perceived academic value and enabling conditions than by social pressure or habitual use in an underexplored secondary-school context. Theoretically, the findings refine the interpretation of UTAUT2 by showing that not all established predictors operate with equal salience in early-stage GenAI adoption. Practically, they suggest that schools should prioritize clear learning purposes, reliable access, technical support, and guided use rather than relying primarily on general promotion of GenAI tools.