Time Series Model for Predicting Dropout in Massive Open Online Courses

Cui Tang, Yuanxin Ouyang, Wenge Rong, Jingshuai Zhang, Zhang Xiong

Abstract: MOOCs are playing an increasing important role in modern education, but the problem of high dropout rate is quite serious. Predicting users’ dropout behavior is an important research direction of MOOCs. In this paper, we extract some raw features from MOOCs uses’ logs and apply the MOOCs users’ daily activities into a recurrent neural network (RNN) with long short-term memory (LSTM) cells, viewing this problem as a time series problem. We collect rich MOOCs users’ log information from XuetangX to test the time series model predicting course drop out. The experiments results indicate that the time series model perform better than other contrast models.