EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Interpretable Undergraduate Dropout Prediction via Explainable Boosting and Gaussian Hidden Markov Models

Edson José Pacheco, Carlos Marcelo Pedroso

IEEE Access · 2026 · 14 · 79029-79044

DOI: 10.1109/access.2026.3696589 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Edson José Pacheco und 1 weiteren Autor:innen. Die Arbeit nennt 47 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Coordenação de Aperfeiçoamento de Pessoal de Nível Superior—Brasil, Article Processing Charge (APC) funded through the CAPES/IEEE Open Access Agreement.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Student dropout remains a persistent challenge in higher education, particularly within engineering and technology undergraduate programs. Although several machine learning approaches achieve high predictive accuracy, their inherent lack of interpretability often constrains pedagogical teams from designing and justifying evidence-based interventions. To address this issue, this paper presents a model that integrates Explainable Boosting Machines (EBM) for cross-sectional risk scoring with Gaussian Hidden Markov Models (G-HMM) for temporal trajectory modeling. The EBM component produces inherently interpretable shape functions to quantify nonlinear relationships between academic indicators and dropout risk. Complementarily, the G-HMM captures transition dynamics by modeling the temporal evolution of these scores across five latent states. The experimental evaluation was conducted using a 15-year dataset from an electrical engineering program. Starting from an initial set of 71 academic indicators, a multi-phase selection algorithm identifies the most discriminative features for each program, ensuring model parsimony and performance. Based on these selected features, the EBM achieved an F1-score of 97.81% with 100% sensitivity. The G-HMM complemented this result by modeling temporal trajectories across five latent states, achieving a mean detection lead time of 2.60 semesters prior to dropout. Unlike traditional baselines, the proposed methodology generates interpretable early-warning signals that facilitate the design and monitoring of targeted interventions. This approach enables institutions to quantify the impact of corrective actions across successive academic cycles.

Förderung

  • Coordenação de Aperfeiçoamento de Pessoal de Nível Superior—Brasil
  • Article Processing Charge (APC) funded through the CAPES/IEEE Open Access Agreement

Zitieren

APA

Pacheco, E. J., & Pedroso, C. M. (2026). Interpretable Undergraduate Dropout Prediction via Explainable Boosting and Gaussian Hidden Markov Models. IEEE Access, 14, 79029-79044. https://doi.org/10.1109/access.2026.3696589

Vancouver

Pacheco EJ, Pedroso CM. Interpretable Undergraduate Dropout Prediction via Explainable Boosting and Gaussian Hidden Markov Models. IEEE Access. 2026;14:79029-79044. doi:10.1109/access.2026.3696589

BibTeX
@article{Pacheco2026,
  author = {Pacheco, Edson José and Pedroso, Carlos Marcelo},
  title = {Interpretable Undergraduate Dropout Prediction via Explainable Boosting and Gaussian Hidden Markov Models},
  journal = {IEEE Access},
  year = {2026},
  volume = {14},
  pages = {79029--79044},
  doi = {10.1109/access.2026.3696589}
}
Kostenlos starten →