Scholarlink

<p>The growth and democratisation of higher education have led to an explosion in the volume of diverse student data being generated, including coursework scores, attendance rates, and behavioural engagement measures. Current enterprise resource planning (ERP) and learning management systems (LMS) are siloed, lack intelligence, and fail to provide real-time, actionable insights. The Intelligent Student Information Management Platform (ISIMP) is a new multi-layered architecture that integrates a data integration layer, a machine-learning-informed processing layer, an interactive application layer, and a secure communication layer into a coherent system. Its distinguishing feature is a closed operational loop in which prediction, risk detection, and intervention are executed inside the production platform rather than as offline analyses. ISIMP uses an ensemble approach that integrates random forest, long short-term memory (LSTM) networks, and gradient-boosted decision trees (GBDTs) to predict student performance, identify at-risk students, and make recommendations. The system was evaluated on a fully specified, seed-controlled synthetic cohort of 12,450 student records (eight weekly snapshots of 23 features per student), released together with all code, saved predictions, and result tables as a reproducibility package. On the held-out test set, the ensemble identifies at-risk students with 88.9% accuracy, 79.1% recall, a 71.4% F1-score, and an AUC-ROC of 0.940—statistically indistinguishable from the strongest individual baselines (logistic regression and XGBoost) and significantly more accurate than a Transformer baseline (McNemar <i>p</i> = 0.0014). The hybrid recommendation engine achieves a higher MAP@5 than pure collaborative filtering (0.202 vs 0.169), with a 78.3% hit rate. An 8.3-day-to-2.7-h (98.6% reduction) mean response time from detection to intervention is achieved, while the assumed 4-week intervention coverage increases from 71.2 to 98.3% (a 38% relative increase) in a scenario-based simulation of administrative workflows. The results were obtained under strictly controlled synthetic conditions and do not demonstrate effectiveness in the field; rather, they prove the feasibility and exact reproducibility of the integrated platform, a necessary step before putting the platform into operation and one that requires validation on real institutional data.</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

Design and implementation of an information management platform for university students: a new model of intelligent education management

  • 期刊论文

  • 发表日期：

    2026年7月31日

查看全文
  • Zheng Qin,
  • Haiyan Wang,
  • Haoyang Fan

摘要

The growth and democratisation of higher education have led to an explosion in the volume of diverse student data being generated, including coursework scores, attendance rates, and behavioural engagement measures. Current enterprise resource planning (ERP) and learning management systems (LMS) are siloed, lack intelligence, and fail to provide real-time, actionable insights. The Intelligent Student Information Management Platform (ISIMP) is a new multi-layered architecture that integrates a data integration layer, a machine-learning-informed processing layer, an interactive application layer, and a secure communication layer into a coherent system. Its distinguishing feature is a closed operational loop in which prediction, risk detection, and intervention are executed inside the production platform rather than as offline analyses. ISIMP uses an ensemble approach that integrates random forest, long short-term memory (LSTM) networks, and gradient-boosted decision trees (GBDTs) to predict student performance, identify at-risk students, and make recommendations. The system was evaluated on a fully specified, seed-controlled synthetic cohort of 12,450 student records (eight weekly snapshots of 23 features per student), released together with all code, saved predictions, and result tables as a reproducibility package. On the held-out test set, the ensemble identifies at-risk students with 88.9% accuracy, 79.1% recall, a 71.4% F1-score, and an AUC-ROC of 0.940—statistically indistinguishable from the strongest individual baselines (logistic regression and XGBoost) and significantly more accurate than a Transformer baseline (McNemar p = 0.0014). The hybrid recommendation engine achieves a higher MAP@5 than pure collaborative filtering (0.202 vs 0.169), with a 78.3% hit rate. An 8.3-day-to-2.7-h (98.6% reduction) mean response time from detection to intervention is achieved, while the assumed 4-week intervention coverage increases from 71.2 to 98.3% (a 38% relative increase) in a scenario-based simulation of administrative workflows. The results were obtained under strictly controlled synthetic conditions and do not demonstrate effectiveness in the field; rather, they prove the feasibility and exact reproducibility of the integrated platform, a necessary step before putting the platform into operation and one that requires validation on real institutional data.