Scholarlink

<p>Machine learning (ML) has advanced educational outcome prediction, yet most systems overlook socioeconomic disparities and long-term model reliability. This study introduces a data-centric Machine Learning Operations (MLOps) framework for large-scale academic forecasting in public high schools, with a specific emphasis on socioeconomic context (SES). The framework targets two tasks: (1) predicting Grade 12 subject scores and (2) forecasting national graduation exam outcomes. A longitudinal dataset of 2283 high school students is used, enriched with a four-level SES attribute: Normal, Poor, Orphaned, and Poor&amp;Orphaned. This socioeconomic indicator enables group-aware data mining and fairness evaluation. Three predictive models–Linear Regression, Multi-Layer Perceptron, and Long Short-Term Memory (LSTM)–were benchmarked under both balanced and imbalanced conditions. Our modular MLOps framework automates data drift detection, retraining, and group monitoring. Data mining results show that incorporating the socioeconomic feature and applying data balancing significantly improves prediction accuracy for disadvantaged students. LSTM models effectively capture non-linear progressions such as late-stage academic surges. While grounded in a Vietnamese public high school, the framework serves as a methodological blueprint for socially responsible, sustainable AI in resource-constrained educational systems worldwide. By embedding socioeconomic awareness into both the modeling and deployment pipeline, this framework enhances predictive performance, fairness, and adaptability. It offers a scalable, interpretable, and open-source solution for educational policymakers seeking equity-driven AI tools.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Data-Centric MLOps Framework for Academic Prediction with Socioeconomic Context in Vietnamese High Schools

  • 期刊论文

  • 发表日期：

    2026年1月9日

查看全文
  • T. T. H. Pham,
  • N. T. K. Oanh,
  • Minh Thuan Pham,
  • T. H. Phuoc Nguyen,
  • Phat-Vo Le Thanh

摘要

Machine learning (ML) has advanced educational outcome prediction, yet most systems overlook socioeconomic disparities and long-term model reliability. This study introduces a data-centric Machine Learning Operations (MLOps) framework for large-scale academic forecasting in public high schools, with a specific emphasis on socioeconomic context (SES). The framework targets two tasks: (1) predicting Grade 12 subject scores and (2) forecasting national graduation exam outcomes. A longitudinal dataset of 2283 high school students is used, enriched with a four-level SES attribute: Normal, Poor, Orphaned, and Poor&Orphaned. This socioeconomic indicator enables group-aware data mining and fairness evaluation. Three predictive models–Linear Regression, Multi-Layer Perceptron, and Long Short-Term Memory (LSTM)–were benchmarked under both balanced and imbalanced conditions. Our modular MLOps framework automates data drift detection, retraining, and group monitoring. Data mining results show that incorporating the socioeconomic feature and applying data balancing significantly improves prediction accuracy for disadvantaged students. LSTM models effectively capture non-linear progressions such as late-stage academic surges. While grounded in a Vietnamese public high school, the framework serves as a methodological blueprint for socially responsible, sustainable AI in resource-constrained educational systems worldwide. By embedding socioeconomic awareness into both the modeling and deployment pipeline, this framework enhances predictive performance, fairness, and adaptability. It offers a scalable, interpretable, and open-source solution for educational policymakers seeking equity-driven AI tools.