Predictive models of academic performance for early intervention in computer programming courses

Low academic performance in computer programming courses is a recurring problem in higher education institutions and has become a significant challenge in the training of future engineers. This challenge is associated with factors such as a lack of prior programming experience, inadequate learning styles, and low motivation, which negatively impact students’ educational progress, compromising both their academic success and profesional competitiveness. This thesis proposes an early intervention system based on academic performance prediction, featuring two types of interventions: preventive and proactive. These interventions provide personalized strategies to support at-risk students, enhance their academic outcomes, and foster self-regulation and motivation in learning programming. The research addresses the question: How do predictions of student performance support early intervention in computer programming courses? To answer this question, a predictive regression model and a classification model were developed using features related to programming activities and grades, with the goal of identifying students at risk of por academic performance during the first weeks of the course. Additionally, intervention strategies were designed and tested, including group tutoring, personalized suggestions, reference source codes, and reinforcement workshops, all aimed at improving students' academic performance. Furthermore, a quasi-experimental study was conducted, and the MSLQ-Colombia questionnaire was applied to assess the effectiveness of the implemented interventions, as well as the learning strategies and motivation employed by students in computer programming courses. As a result of this thesis, eight articles were written, significantly contributing to the achievement of the research objectives and the advancement of the field of Computer Science. Among these, three were published in Q1-ranked journals, two in Q3-ranked journals, and the remaining three are currently under review. The first set of articles addresses the predictive models developed and a pedagogical model based on Kirkpatrick's framework, designed to measure the effectiveness of teaching strategies in programming courses. Subsequent works analyze how the flipped classroom approach can be effectively integrated into early intervention systems in these courses. The remaining documents include a literature review on Early Warning Systems and their contribution to predicting academic performance, as well as the early intervention tool developed and the quasi-experimental study conducted. Among the most significant research findings is the effectiveness of the predictive models implemented, which achieved up to 96% accuracy in predicting academic performance during the first weeks of the course. The early intervention tool developed allowed for the identification of at-risk students and the provision of personalized interventions. These results demonstrate that early interventions, supported by academic predictions, are essential for improving both the performance and retention of students in computer programming courses.

• Autores: Llanos Mosquera, José Miguel
• Tipo de recurso: Doctoral thesis
• Fecha de publicación: 2024
• Institución: Universidad del Valle
• Repositorio: Repositorio Digital Univalle
• Idioma: eng
• OAI Identifier: oai:bibliotecadigital.univalle.edu.co:10893/36049
• Acceso en línea: https://hdl.handle.net/10893/36049
• Palabra clave: Programación (Informática); Ciencias de la computación; Código fuente; Algoritmos de aprendizaje automático; Estrategias de aprendizaje
• Rights: openAccess
• License: https://creativecommons.org/licenses/by-nc-nd/4.0/