Scholarlink

<p>Effective teaching of college-level Data Structures is critical for improving students’ computational thinking and problem-solving skills. But traditional static course designs often fail to accommodate diverse learning needs, resulting in suboptimal engagement and knowledge retention. Addressing this challenge, this research proposes an adaptive curriculum optimization framework using Weighted Pelican-Double Deep Q Network (WP-Double DQN) to dynamically design and sequence course content. 2000 student performance data were collected from open sources, including Learning Management Systems (LMS), quiz scores, assignment submissions, and classroom interaction logs. To ensure high-quality, constant data for the learning optimization process, missing value imputation and normalization were applied. Subsequently, Principal Component Analysis (PCA) was utilized for feature extraction, reducing dimensionality while retaining key indicators such as comprehension, problem-solving ability, and engagement trends. The WP-Double DQN agent models curriculum design as a sequential decision-making process, selecting and prioritizing modules to maximize learning rewards. Its weighted double Q-learning reduces overestimation errors, ensuring accurate evaluation of module impact, while the Pelican-inspired weighting emphasizes high-impact modules, directly supporting the objective of delivering a personalized, adaptive, and effective learning experience that enhances student understanding and engagement. Experimental results are evaluated by Python 3.10, and they demonstrate that the proposed framework significantly predicts improvement of student outcomes, achieving 95.4% accuracy, 93.2% recall, 94.3% F1-score, 94.4% precision, 0.12 MAE, 0.15 RMSE, and 0.96 AUC-ROC. This research provides a novel Deep Reinforcement Learning (DRL)-driven methodology for curriculum optimization, highlighting the potential of WP-Double DQN for intelligent, data-informed educational design and broader applications in adaptive learning systems.</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

Deep reinforcement learning-based optimization design for college data structures course content

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月30日

查看全文
  • Dongjuan Che,
  • Jianwei Zhou

摘要

Effective teaching of college-level Data Structures is critical for improving students’ computational thinking and problem-solving skills. But traditional static course designs often fail to accommodate diverse learning needs, resulting in suboptimal engagement and knowledge retention. Addressing this challenge, this research proposes an adaptive curriculum optimization framework using Weighted Pelican-Double Deep Q Network (WP-Double DQN) to dynamically design and sequence course content. 2000 student performance data were collected from open sources, including Learning Management Systems (LMS), quiz scores, assignment submissions, and classroom interaction logs. To ensure high-quality, constant data for the learning optimization process, missing value imputation and normalization were applied. Subsequently, Principal Component Analysis (PCA) was utilized for feature extraction, reducing dimensionality while retaining key indicators such as comprehension, problem-solving ability, and engagement trends. The WP-Double DQN agent models curriculum design as a sequential decision-making process, selecting and prioritizing modules to maximize learning rewards. Its weighted double Q-learning reduces overestimation errors, ensuring accurate evaluation of module impact, while the Pelican-inspired weighting emphasizes high-impact modules, directly supporting the objective of delivering a personalized, adaptive, and effective learning experience that enhances student understanding and engagement. Experimental results are evaluated by Python 3.10, and they demonstrate that the proposed framework significantly predicts improvement of student outcomes, achieving 95.4% accuracy, 93.2% recall, 94.3% F1-score, 94.4% precision, 0.12 MAE, 0.15 RMSE, and 0.96 AUC-ROC. This research provides a novel Deep Reinforcement Learning (DRL)-driven methodology for curriculum optimization, highlighting the potential of WP-Double DQN for intelligent, data-informed educational design and broader applications in adaptive learning systems.