Improving Prediction of Student Performance based on Multiple Feature Selection Approaches

Recently, to provide the better education for students, there are a lot of researchers that discover the latent characteristics of students for predicting their performance. However, few existing work has explored the problem of extracting information from E-learning to get more precise and interpretable analysis. Based on the edX open data, we first predict whether the student will obtain the certificate, and take it as the criterion of student perfomance. Next, according to the requirement of the research, student features on dataset can be classified into three categories primarily, and some characters which seem umimportant intuitively have been removed already. Then, we adopt several kinds of feature selection approaches to extract important influencing student feature of the rest characters. Finally, though a few of existing classical machine learning methods, we build the model and predict student performance. The extensive experiments on the dataset of edX open platform we have conducted validated the effectiveness of predicting student performance.