Dataset descriptions.

Student performance is crucial for addressing learning process problems and is also an important factor in measuring learning outcomes. The ability to improve educational systems using data knowledge has driven the development of the field of educational data mining research. Here, this paper proposes a machine learning method for the prediction of student performance based on online learning. The critical thought is that eleven learning behavioral indicators are constructed according to online learning process, following that, through analyzing the correlation between the eleven learning behavioral indicators and the scores obtained by students online learning, we filter out those learning behavioral indicators that are weakly correlated with student scores, meanwhile, retain these learning behavior indicators being strongly correlated with student scores, which are used as the eigenvalue indicators. Finally, using the eigenvalue indicators to train the proposed logistic regress model with Taylor expansion. Experimental results show that the proposed logistic regress model defeats against the comparative models in prediction ability. Results also indicate that there is a significant dependency between students’ initiative in learning and learning duration, nevertheless, learning duration has a significant effect on the prediction of student performance.学生表现对于解决学习过程中的问题至关重要，同时也是衡量学习成果的核心指标。依托数据知识优化教育系统的需求，推动了教育数据挖掘（Educational Data Mining）研究领域的发展。本文提出一种基于在线学习（Online Learning）的机器学习（Machine Learning）方法用于学生表现预测，其核心思路为：根据在线学习流程构建11项学习行为指标；随后，通过分析这11项学习行为指标与学生在线学习所得成绩的相关性，筛除与学生成绩相关性较弱的学习行为指标，保留与学生成绩相关性较强的指标作为特征指标；最后，利用该特征指标训练本文提出的基于泰勒展开（Taylor Expansion）的逻辑回归（Logistic Regression）模型。实验结果表明，本文提出的逻辑回归模型在预测性能上优于对比模型。研究结果同时显示，学生的学习主动性与学习时长之间存在显著依存关系，且学习时长对学生表现预测具有显著影响。