Classification of Student Profiles Based on Generative AI Dependency and Complex Problem-Solving Capacity: A Machine Learning Approach Using K-Means Clustering

Abstract

The rapid integration of Generative Artifi cial Intelligence (GAI) tools, particularly ChatGPT, into higher education has raised concerns about potential cognitive dependency and its impact on students' problem-solving skills. This study aims to classify student profi les based on their dependency on generative AI and their ability to solve complex problems using machine learning clustering techniques. A cross-sectional study was conducted with 847 engineering students from three Latin American universities. The AI Dependency Scale for University Students (EDIAU-15), the Complex Problem-Solving Inventory (IRPC-25), and the ChatGPT Frequency of Use Questionnaire (CFUC) were administered. K-means clustering with Davies-Bouldin optimization identifi ed four distinct student profi les: (1) Critical Autonomists (23.4%, n=198): low AI dependence, high problem-solving ability; (2) Balanced Integrators (31.2%, n=264): moderate AI use, preserved cognitive skills; (3) Dependent Compensators (28.6%, n=242): high dependence on AI, declining problem-solving ability; and (4) Passive Delegators (16.8%, n=143): severe dependence on AI, signifi cantly impaired cognitive functioning. ANOVA revealed signifi cant diff erences between profi les in academic performance (F=47.82, p<.001), metacognitive awareness (F=38.94, p<.001), and self-effi cacy (F=52.17, p<.001). Structural equation modeling confi rmed that AI dependence mediates the relationship between ChatGPT usage frequency and problem-solving ability (β=-0.43, p<.001).