Título: A New Machine Learning Approach to Detect Student Success in Pair Programming
Autores: Gabriela Bezerra, Felipe Pinto Marinho & Ajalmar Rêgo da Rocha Neto
Resumo: Learning software development remains a significant challenge for students, even when supported by active learning methodologies such as Challenge-Based Learning (CBL). In CBL, students often work in pairs or groups to develop applications that address real-world problems. However, interpersonal issues and varying collaboration styles can negatively affect performance. This study proposes a ML approach to predicting which student pairs are likely to struggle, based on social styles (using the Merrill-Reid model) and project complexity measurements. A dataset of 38 paired programming projects was collected from a real classroom setting. Supervised machine learning models — Support Vector Machine (SVM), k-Nearest Neighbors (KNN), and Decision Tree — were trained and evaluated using the Leave-One-Out methodology, with model tuning done via grid search and cross-validation. The best-performing model (KNN) achieved an accuracy of 78.94% in identifying pairs that would meet at least 70% of the project requirements. The results demonstrate the potential of data-efficient models to support timely instructional interventions in collaborative learning environments.
Palavras-chave: education; software development; pair programming; challenge based learning; machine learning; data-efficient.
Páginas: 8
Código DOI: 10.21528/CBIC2025-1175362
Artigo em PDF: CBIC_2025_paper1175362.pdf
Arquivo BibTeX:
CBIC_2025_1175362.bib