EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

VertiDaX: an explainable hybrid deep learning model for dropout prediction with vertical integration

Mohammad Waseem, Shafiqul Abidin, Mahfooz Alam, Mohd Shahid Husain, Mohammad Zunnun Khan, Zubair Ashraf

PeerJ Computer Science · 20.03.2026 · 12 · e3718

DOI: 10.7717/peerj-cs.3718 ↗

Kurz gefasst

Fachartikel in der Zeitschrift PeerJ Computer Science, online erschienen am 20.03.2026. Verfasst von Mohammad Waseem und 5 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 38 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Deanship of Graduate Studies and Scientific Research at University of Bisha, Fast-Track Research Support Program.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Massive Open Online Courses (MOOCs) offer access to high-scalable and flexible education to diverse people in distant parts of the world, yet persistently plagued by high dropout rates. To operationalize this problem, more correct and interpretable models need to be created, which operate on heterogeneous and vertically integrated educational data. Here, the method is a hybrid explainable deep learning model, used to predict Dropout in MOOCs. The student activity logs, course metadata, and assessment records used in the framework are combined through integration with an unbalanced assignment to ensure that imbalanced data are balanced out with the application of Generative Adversarial Network (GAN). To facilitate complex behavioral patterns related to Dropout, we come up with a multi-head attention neural network that represents both temporal and contextual aspects. To provide interpretability, we acquire local explanations with Local Interpretable Model-agnostic Explanations (LIME) and global cohort-level representativeness on the basis of the attention mechanism. The combination of these explanations is carried out by a new hybrid algorithm that offers a population level reformulation as well as a student-level ratification of predictions. On the benchmark Knowledge Discovery and Data Mining (KDD Cup) 2015 dataset, the proposed VertiDaX gives an excellent F1-score and receiver operating characteristic area under the curve (ROC-AUC), which are better than earlier baselines. The study also used the Open University Learning Analytics Dataset (OULAD) data to stratify cross validate on the same label scheme, where we evaluate generalisation and stability in moderate class imbalance. The explained method captures common and personalized behavioral patterns associated with dropout, with a view to establishing educator-specific intervention strategies. This project represents a stepping stone in showing that GAN-based data augmentation plus attention-based deep learning and hybrid interpretation is a formula that will offer valuable and concrete results in predicting dropout rates in MOOCs. We propose the exciting future use of interpretable artificial intelligence (AI) that can be implemented in scalable online training based on the findings.

Förderung

  • Deanship of Graduate Studies and Scientific Research at University of Bisha
  • Fast-Track Research Support Program

Zitieren

APA

Waseem, M., Abidin, S., Alam, M., Husain, M. S., Khan, M. Z., & Ashraf, Z. (2026). VertiDaX: an explainable hybrid deep learning model for dropout prediction with vertical integration. PeerJ Computer Science, 12, e3718. https://doi.org/10.7717/peerj-cs.3718

Vancouver

Waseem M, Abidin S, Alam M, Husain MS, Khan MZ, Ashraf Z. VertiDaX: an explainable hybrid deep learning model for dropout prediction with vertical integration. PeerJ Computer Science. 2026;12:e3718. doi:10.7717/peerj-cs.3718

BibTeX
@article{Waseem2026,
  author = {Waseem, Mohammad and Abidin, Shafiqul and Alam, Mahfooz and Husain, Mohd Shahid and Khan, Mohammad Zunnun and Ashraf, Zubair},
  title = {VertiDaX: an explainable hybrid deep learning model for dropout prediction with vertical integration},
  journal = {PeerJ Computer Science},
  year = {2026},
  volume = {12},
  pages = {e3718},
  doi = {10.7717/peerj-cs.3718}
}
Kostenlos starten →