ABSTRACT

On e-learning platforms, most e-learners didn’t complete the course successfully. It means that reducing dropout is a critical problem for the sustainability of e-learning. This paper aims to establish a predictive model to describe e-learners’ dropout behavior, which can help the commercial e-learning platforms to make appropriate interventions and incentives. First of all, we defined the features of unusual learning behaviors in commercial e-learning platform, and used the Cox proportional hazard model of survival analysis to select variables that can reasonably predict dropout possibilities. Results show that there are six variables which have significant influence on dropout behavior: dropout history, number of watched videos, number of progress bar operation, number of test questions operation, number of weeks that the login frequency is higher than average, and payment status. We also proposed cumulative gain, predicted retention number and predicted dropout learner number in next period, to evaluate the application ability of the predictive model. Finally, we performed an empirical analysis and verified the predictive effectiveness. The further application of the predictive model also shows that it can help the managers of e-learning platforms to adjust their strategy to improve the retention rate of potential lost learners.

摘要

在在线学习平台上，大多数在线学习者都没有成功完成课程。这意味着减少辍学是电子学习可持续性的关键问题。本文旨在建立一个预测模型来描述在线学习者的辍学行为，以帮助商业在线学习平台做出适当的干预和激励。首先，我们定义了商业网络学习平台异常学习行为的特征，并利用生存分析的Cox比例风险模型来选择能够合理预测辍学可能性的变量。结果表明，有六个变量对dropout行为有显着影响：dropout历史、视频观看次数、进度条操作次数、试题操作次数、登录频率高于平均水平的周数，以及付款状态。我们还提出了累积增益、预测的保留数和预测的下一个时期的辍学学习者数，以评估预测模型的应用能力。最后，我们进行了实证分析，验证了预测的有效性。预测模型的进一步应用也表明，它可以帮助在线学习平台的管理者调整策略，提高潜在流失学习者的保留率。