Abstract:

Understanding student diversity in undergraduate research experiences remains critical for personalized educational interventions. This study developed an interpretable machine learning pipeline to identify and classify distinct motivational profiles among 500 programming students in Colombian undergraduate research programs. Data were collected through a 35-item survey (2021-2025) capturing motivational, behavioral, and attitudinal dimensions. Five clustering algorithms were evaluated, with Gaussian Mixture Models (GMM) selected as optimal (Silhouette Score=0.146), identifying three profiles: Academic Improvers (knowledge-focused), Vocational Explorers (hands-on career exploration), and Employability-Oriented (job market preparation). Six classification algorithms were compared, with Multi-Layer Perceptron achieving 94% accuracy (F1-Score=0.94). SHapley Additive exPlanations (SHAP) analysis revealed intrinsic learning motivation as the primary discriminator (SHAP importance: 0.352 for Academic Improvers, 0.365 for Employability-Oriented, 0.019 for Vocational Explorers), while frontend development interest distinguished Vocational Explorers (0.128). The high classification accuracy and model interpretability demonstrated feasibility for automated student profiling in educational systems. These findings contributed to explainable AI in education, providing actionable insights for curriculum design and student support services. Future research should validate findings across diverse contexts and integrate behavioral analytics for enhanced predictive accuracy.

Keywords:

Explainable Artificial Intelligence, Educational Data Mining, Student Profiling, Machine Learning Classification, Undergraduate Research Experiences.