An intelligent tool for early drop-out prediction of distance learning students

Early identification of vulnerable students who are prone to drop-out is critical for devising effective educational retention strategies. Based on the Activity Theory, we undertake this challenge by considering students’ online activities as a useful predictor of their academic performance. Specifically, six artificial intelligence and related prediction models in individual and ensemble structures for tackling classification and multi-objective optimization tasks pertaining to early prediction of students’ performance are presented. A real database comprising online learning activities of 2544 students over 2 years in 84 science, engineering, and technology courses from an open distance education institution is used for evaluation. Comparing with other studies in the literature, the huge numbers of students and courses involved in this study pose a great challenge, due to increase in complexity of the problem and data dimensionality. The empirical results reveal statistically significant improvements of the ensemble-based models as compared with individual models in prediction of students’ performance. Implications of the results are analyzed and discussed from the Activity Theory perspective.