Title: Non-Academic Interests and Competency Prediction: A Proof-of-Concept Multi-Reference ML Pipeline for Student Profiling
Authors: Pham Ngoc Thanh, Nguyen Khanh Toan, Trinh Thanh Do
Volume: 10
Issue: 9
Pages: 228-236
Publication Date: 2026/09/28
Abstract:
Existing models for academic prediction oversimplify non-academic interests and rely on single data sources. This proof-of-concept presents a multi-reference machine-learning pipeline predicting four student competencies using Likert-scale interest profiles from four distinct sources. Evaluated on a reproducible synthetic cohort (n=390), the protocol tests feature ablation, three source-fusion architectures, inter-source disagreement, and SHAP explanation stability. Results under this generative baseline were predominantly null: adding interest features, multiple sources, or disagreement metrics yielded no significant predictive improvement over an academic baseline. Furthermore, gated weighting proved the weakest fusion strategy, and SHAP rankings lacked stability in complex configurations. Ultimately, this study contributes a fully specified, end-to-end reproducible analysis protocol ready for empirical application, rather than a deployable predictive model.