A MULTIVARIATE TIME SERIES PREDICTION MODEL OF ACADEMIC PERFORMANCE BASED ON LSTM-TRANSFORMER-SVM
DYNA · 01.11.2026 · 101 (6) · 517-524
DOI: 10.52152/d11557 ↗Kurz gefasst
Fachartikel in der Zeitschrift DYNA, online erschienen am 01.11.2026. Verfasst von Jun Zou und 4 weiteren Autor:innen.
Abstract (Originalsprache)
With the rapid development of intelligent education research, academic performance prediction based on multi-period achievement data has become a research focus. To address the problems that traditional methods have difficulty capturing long-term dependencies and complex nonlinear features, and have limited prediction accuracy and generalization ability under small-sample conditions, this study proposes a hybrid prediction model integrating LSTM, SVM, and Transformer. This model combines the time series modeling ability of LSTM, the self-attention feature extraction ability of Transformer, and the nonlinear regression ability of SVM to improve the accuracy and stability of academic performance prediction. Taking the scores of four examinations in a statistics course at a university as samples, the training set, validation set, and test set were divided at a ratio of 8:1:1, and Bayesian optimization and particle swarm optimization algorithms were introduced in the later stage for hyperparameter tuning. By comparing the LSTM+SVM+Transformer model with single models, including LSTM, SVM, and Transformer, and other hybrid models, including LSTM+Transformer, SVM+Transformer, and LSTM+SVM, the experimental results show that the proposed model is significantly superior to other methods in terms of prediction accuracy and robustness. In addition, SHAP analysis was used to evaluate the interpretability of the model, and the results show that recent examination scores make the greatest contribution to the prediction of final grades, providing a quantitative basis for personalized teaching intervention and educational management decision-making. Keywords: Time-series-based academic performance prediction; Long Short-Term Memory; Transformer; Support Vector Machine
Zitieren
APA
Zou, J., Guo, W., Wang, X., Wu, Z., & DU, J. (2026). A MULTIVARIATE TIME SERIES PREDICTION MODEL OF ACADEMIC PERFORMANCE BASED ON LSTM-TRANSFORMER-SVM. DYNA, 101(6), 517-524. https://doi.org/10.52152/d11557
Vancouver
Zou J, Guo W, Wang X, Wu Z, DU J. A MULTIVARIATE TIME SERIES PREDICTION MODEL OF ACADEMIC PERFORMANCE BASED ON LSTM-TRANSFORMER-SVM. DYNA. 2026;101(6):517-524. doi:10.52152/d11557
BibTeX
@article{Zou2026,
author = {Zou, Jun and Guo, Weiting and Wang, Xiaofei and Wu, Zhaoyu and DU, JUAN},
title = {A MULTIVARIATE TIME SERIES PREDICTION MODEL OF ACADEMIC PERFORMANCE BASED ON LSTM-TRANSFORMER-SVM},
journal = {DYNA},
year = {2026},
volume = {101},
number = {6},
pages = {517--524},
doi = {10.52152/d11557}
}
EUVIMED