An Interpretable Cart-Based Framework for Multi-Target Educational Prediction Using Feature Selection and Model Pruning

Abstract

The implementation of the K-12 Basic Education Program in the Philippines introduced Senior High School (SHS) as a critical stage where students must select an academic strand aligned with their interests, abilities and future career goals. This decision is particularly significant because it influences students’ academic readiness, motivation, and long-term educational outcomes. For Grade 10 learners, choosing among SHS strands such as Science, Technology, Engineering, and Mathematics (STEM), Accountancy, Business, and Management (ABM), Humanities and Social Sciences (HUMSS), and General Academic Strand (GAS) often occurs at a formative stage when self-awareness and academic guidance are still developing. Consequently, inaccurate or poorly informed strand choices may lead to academic difficulties, disengagement, or later program shifts, underscoring the importance of informed and evidence-based SHS decision making.
In recent years, educational institutions have increasingly explored the use of machine learning (ML) techniques to support academic advising and student performance prediction. ML-based decision-support systems offer the potential to analyse large volumes of student data and uncover patterns that may not be immediately apparent through traditional counselling approaches. However, despite their predictive power, many existing ML models particularly ensemble and deep learning methods operate as black-box systems. These models generate predictions without providing clear explanations of how decisions are made, limiting their suitability for educational contexts where transparency, accountability, and human oversight are essential.