Abstract
Selecting important features in an e-learning environment is crucial for predicting student academic performance. E-learning offers personalized and uninterrupted connections and communication between students and other learning contexts. The increasing proliferation of smart technologies has enabled students to acquire and connect to learning materials and instruction anytime, anywhere. Obviously, the student's interaction behaviors in e-learning environment have been widely considered. In fact, the interaction in the e-learning system and its impact on students' performance is subject to discussion and interest. This study, for the most part, focusses on two targets: the first is to find critical factors that affect student's outcomes in the e-learning system for illustration and the second is more tied, building a well-performed prediction model. The main contribution is twofold: to highlight some experimental visions in the influence of a set of variables using features selection techniques and to propose a prediction model involving the most relevant features applying K-fold Cross Validation method. Different variables effect on model performance and correlations between the input and the target output are discussed in detail using student data provided by the Learning Management System. The recommended method is, then, compared with another popular machine learning methods. The results exposed that, the student with greater engagement in the e-learning system leads to significantly higher performance; however, students who get low in the course tend to interact less frequently. Furthermore, study results indicate that some prediction techniques such as the Random Forest method have considerable advantages in student performance prediction reached up to 80% of accuracy. Other students' features that may be effective in the e-learning system are also discussed.