New Horizons of Science, Technology and Culture Vol. 11

The rapid adoption of artificial intelligence (AI) tools in education has significantly transformed learning practices, offering enhanced support for academic tasks such as writing, problem-solving, and research. However, excessive reliance on these tools may negatively impact students’ critical thinking, originality, and overall academic performance. This study proposes a machine learning-based framework to detect students at academic risk due to excessive AI tool usage by analysing behavioural and performance-related data. The research integrates multidimensional inputs, including AI usage frequency, session duration, assignment similarity scores, attendance, and GPA, to develop predictive models for early risk identification. A comprehensive data analysis process involving exploratory data analysis, feature engineering, and model training was conducted using multiple classifiers, including Logistic Regression, Decision Trees, Support Vector Machines, and Random Forest. Among these, the Random Forest classifier demonstrated superior performance, achieving an accuracy of 88%, a precision of 86%, a recall of 83%, an F1-score of 84%, and an AUC of 0.91. Feature importance analysis revealed that AI usage frequency and assignment similarity scores are the most significant predictors of academic risk, highlighting the critical role of behavioural patterns over traditional academic indicators. Correlation analysis further confirmed a negative relationship between AI usage and academic performance, alongside a positive association with content similarity, indicating potential overdependence on AI-generated outputs. This manuscript addresses a critical and emerging challenge in AI-enabled education by examining the impact of excessive AI tool usage on student performance. It contributes to the scientific community by proposing a novel machine learning-based framework that integrates behavioural analytics for early identification of at-risk students. The study provides empirical evidence that AI usage patterns are strong predictors of academic risk, offering new insights for educational data mining research. Additionally, it supports the development of proactive, data-driven interventions to enhance academic integrity and student success.

The findings emphasise the dual-edged nature of AI in education, where its benefits must be balanced with responsible usage. The proposed framework enables early detection of at-risk students and supports data-driven, proactive intervention strategies to enhance academic integrity and student success. This study contributes to the growing field of educational data mining by demonstrating the effectiveness of machine learning in monitoring emerging learning behaviours in AI-assisted environments.