Identifying At-Risk Students in Higher Education: A Multi-Institutional Study for Data-Driven Intervention and Student Success

Early identification of at-risk students in higher education is crucial for implementing timely interventions and enhancing academic outcomes. Recent advancements in Artificial Intelligence have created opportunities to forecast students’ academic outcomes using different Machine Learning (ML) models. Previous studies have emphasized demographic, academic, social, and environmental factors while contextual and institutional factors remain underexplored, reducing the accuracy of existing models. Prior research relies on single-institution datasets, limiting the generalizability of results. There is also inadequate empirical evidence to support student persistence theories. This study extends beyond traditional factors by examining comprehensive behavioral, psychological, and institutional-contextual influences. The authors have collected data from undergraduate students of ten different public and private sector universities in Pakistan and explored new predictors. Unlike binary or three-class models, this study has classified the predicted results into five letter grades (A-E), leading to more precise and insightful predictions. Different types of feature selection techniques have been employed to determine the most influential factors affecting students’ academic performance, and a comparative analysis is performed through several ML algorithms. The performance parameters have shown that Gradient Boosting with the Recursive Feature Elimination technique outperforms the other ML methods with an accuracy of 0.8122. In addition to traditional parameters, this study has explored four new thematic factors of Temporary Visiting Faculty (TVF), unannounced quizzes, online lectures, and discrimination by teachers, thus offering new insights into the field of students’ performance analysis. These findings can help institutions make data-driven decisions to support student success and guide policy development.