An ideal learning analytics tool for programming exercises performs the role of a lecturer who monitors the code development, provides customized support and identifies students at risk to drop out. But a reliable prediction and prevention of drop-out is difficult, due to the huge problem space in programming tasks and variety of solutions and programming strategies. The purpose of this paper is to tackle this problem by, first, identifying activity patterns that indicate students at risk; and, second, finding reasons behind specific activity pattern, for identification of instructional interventions that prevent drop-out.Design/methodology/approachThe authors combine two investigation strategies: first, learning analytic techniques (decision trees) are applied on features gathered from students, while completing programming exercises, in order to classify predictors for drop-outs. Second, the authors determine cognitive, motivational and demographic learner characteristics based on a questionnaire. Finally, both parts are related with a correlation analysis.FindingsIt was possible to identify generic variables that could predict early and later drop-outs. For students who drop out early, the most relevant variable is the delay time between availability of the assignment and the first login. The correlation analysis indicates a relation with prior programming experience in years and job occupation per week. For students who drop out later in the course, the number of errors within the first assignment is the most relevant predictor, which correlates with prior programming skills.Originality/valueThe findings indicate a relation between activity patterns and learner characteristics. Based on the results, the authors deduce instructional interventions to support students and to prevent drop-outs.
Journal of Applied Research in Higher Education – Emerald Publishing
Published: Jan 17, 2020
Keywords: Motivation; Drop-out; Learning analytics; Programming course