Modeling and machine learning for analyzing and predicting learning activities in robotics education

Interpretable artificial intelligence can support data-driven decisions in robotics education, where instructors must track heterogeneous student progress in complex, practice-oriented environments. This paper investigates how machine learning models can analyze and predict student performance in school and university robotics courses using digital traces from learning management systems, robotics simulators, and assessment logs. We construct a feature set that captures task completion time, frequency of compilation and simulation runs, error patterns in code, participation in teamwork activities, and prior achievement. Several supervised learning algorithms are compared, including gradient-boosted decision trees and interpretable rule-based models. Model explainability is ensured through global feature importance analysis and local explanations that highlight which behaviors most strongly influence each student's predictions. Experiments on datasets collected from introductory robotics courses demonstrate that the proposed approach predicts final achievement and failure risk with high accuracy and stable generalization across cohorts. The interpretable explanations reveal meaningful behavioral profiles, such as students who experiment intensively in the simulator but rarely submit intermediate solutions. These findings provide instructors with actionable insights for early intervention, adaptive feedback, and personalized support, and they illustrate the potential of transparent machine learning tools to improve the design and management of robotics education.

Перевод пока недоступен