EUVIMEDEUVIMED entdecken
Open Access · CC BY-SA 4.0

A Multi-Model AI Pipeline for Student Risk Prediction, Profiling, Knowledge Tracing, and Resource Recommendation in Intelligent Learning Systems

Rui Jiang, Yuan Jiang, Qiaoling Cheng, GuiHong Wu

Journal of Computer Science and Frontier Technologies · 30.10.2026 · 4 (3) · 59

DOI: 10.63313/jcsft.9108 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Computer Science and Frontier Technologies, online erschienen am 30.10.2026. Verfasst von Rui Jiang und 3 weiteren Autor:innen. Die Arbeit nennt 14 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-SA 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

How can a learning platform turn raw click streams and gradebook entries into timely, personalized support for students? This question motivated the work reported here. We describe an end-to-end AI training pipeline for an Intelligent Learning Assistance System (ILAS) that we built to support undergraduate Java programming instruction. Rather than betting on a single modeling paradigm, our pipeline brings together four complementary approaches: XGBoost for academic risk prediction, K-Means for student profiling, Bayesian Knowledge Tracing for knowledge-state estimation, and BERT for semantic encoding of learning resources. Working with behavioral data from 204 students across four classes, we extract a 23-dimensional feature vector from 11 heterogeneous Excel sheets. The XGBoost classifier reaches 90.2% accuracy and 0.993 AUC (one-vs-rest) in a three-class risk task. K-Means clustering, using 11 behavioral features, identifies four interpretable learner profiles (A through D) with a silhouette score of 0.398. We train 15 BKT models via the Baum-Welch EM algorithm and produce a 204 × 15 mastery matrix. A pre-trained bert-base-chinese model encodes 15 learning resources into 768-dimensional vectors using both CLS and mean pooling. Finally, a grid search over 45 weight combinations fuses BKT recall, BERT semantic similarity, and collaborative filtering into a recommendation strategy that achieves Precision@5 of 1.0 in offline evaluation. The entire pipeline completes in about 70 seconds on a GPU and outputs 16 serialized artifacts ready for production use.

Zitieren

APA

Jiang, R., Jiang, Y., Cheng, Q., & Wu, G. (2026). A Multi-Model AI Pipeline for Student Risk Prediction, Profiling, Knowledge Tracing, and Resource Recommendation in Intelligent Learning Systems. Journal of Computer Science and Frontier Technologies, 4(3), 59. https://doi.org/10.63313/jcsft.9108

Vancouver

Jiang R, Jiang Y, Cheng Q, Wu G. A Multi-Model AI Pipeline for Student Risk Prediction, Profiling, Knowledge Tracing, and Resource Recommendation in Intelligent Learning Systems. Journal of Computer Science and Frontier Technologies. 2026;4(3):59. doi:10.63313/jcsft.9108

BibTeX
@article{Jiang2026,
  author = {Jiang, Rui and Jiang, Yuan and Cheng, Qiaoling and Wu, GuiHong},
  title = {A Multi-Model AI Pipeline for Student Risk Prediction, Profiling, Knowledge Tracing, and Resource Recommendation in Intelligent Learning Systems},
  journal = {Journal of Computer Science and Frontier Technologies},
  year = {2026},
  volume = {4},
  number = {3},
  pages = {59},
  doi = {10.63313/jcsft.9108}
}
Kostenlos starten →