Distributed Ensemble Based Iterative Classification for Churn Analysis and Prediction of Dropout Ratio in E-Learning

Churn in e-learning refers to learners who gradually perform less and become lethargic and may potentially drop out from the course. Churn prediction is a highly sensitive and critical task in an e-learning system because inaccurate predictions might cause undesired consequences. A lot of approaches proposed in the literature analyzed and modeled churn prediction using learner's personal attributes from learner profiles and their overall performance. The major concern with existing approaches is that the accuracy of prediction is not satisfactory as the model is built on the sample data with limited features. This paper addresses this issue by proposing a distributed iterative classifier that deploys an ensemble learning algorithm to generalize the model for predicting potential churn from personal attributes. Predictions from the base classifier are obtained using a distributed iterative classification algorithm that deploys a map-reduce framework. Iterative classification algorithm predicts signs of attrition in the learners through their online interactions. It can also process a very large network, which was lacking in the existing solution. The proposed system is evaluated using the features of five students and results are reported. Experimental results show that the proposed ensemble classifier not only improves the performance of churn prediction significantly but also runs faster than the other algorithms.