ABSTRACT E-learning platforms and processes face several challenges, among which is the idea of personalizing the e-learning experience and to keep students motivated and engaged. This work is part of a larger study that aims to tackle these two challenges using a variety of machine learning techniques. To that end, this paper proposes the use of k-means algorithm to cluster students based on 12 engagement metrics divided into two categories: interaction-related and effort-related. Quantitative analysis is performed to identify the students that are not engaged who may need help. Three different clustering models are considered: two-level, three-level, and five-level. The considered dataset is the students’ event log of a second-year undergraduate Science course from a North American university that was given in a blended format. The event log is transformed using MATLAB to generate a new dataset representing the considered metrics. Experimental results’ analysis shows that among the considered interaction-related and effort-related metrics, the number of logins and the average duration to submit assignments are the most representative of the students’ engagement level. Furthermore, using the silhouette coefficient as a performance metric, it is shown that the two-level model offers the best performance in terms of cluster separation. However, the three-level model has a similar performance while better identifying students with low engagement levels.

中文翻译：

---

电子学习环境中的学生参与度：使用 K 均值聚类

摘要 电子学习平台和流程面临着若干挑战，其中包括个性化电子学习体验和保持学生积极性和参与度的想法。这项工作是一项更大研究的一部分，该研究旨在使用各种机器学习技术解决这两个挑战。为此，本文提出使用 k-means 算法基于 12 个参与度指标对学生进行聚类，分为两类：交互相关和努力相关。执行定量分析以识别可能需要帮助的未参与的学生。考虑了三种不同的聚类模型：两级、三级和五级。所考虑的数据集是以混合格式提供的北美大学二年级本科科学课程的学生事件日志。使用 MATLAB 转换事件日志以生成表示所考虑指标的新数据集。实验结果分析表明，在考虑的与交互相关和与努力相关的指标中，登录次数和提交作业的平均持续时间最能代表学生的参与度。此外，使用轮廓系数作为性能指标，表明两级模型在聚类分离方面提供了最佳性能。然而，三级模型具有相似的性能，同时更好地识别低参与度的学生。登录次数和提交作业的平均时长最能代表学生的参与度。此外，使用轮廓系数作为性能指标，表明两级模型在聚类分离方面提供了最佳性能。然而，三级模型具有相似的性能，同时更好地识别低参与度的学生。登录次数和提交作业的平均时长最能代表学生的参与度。此外，使用轮廓系数作为性能指标，表明两级模型在聚类分离方面提供了最佳性能。然而，三级模型具有相似的性能，同时更好地识别低参与度的学生。