Construction of Online Distance Learning Risk Prediction Model Based on Data Mining

As distance learning becomes more popular, it is increasingly important to model learner learning risk predictions in order to reduce the risk of failing a course and motivate learning. We collected 13,000 online students' background information and online learning behavior data of two different courses from a training and continuing education institution in Chongqing. We adopted SMOTE Sampling and Cost-Sensitive Algorithm to solve the problem of uneven distribution of data samples. We used Logistic Regression, Bayesian Network, Neural Network, SVM, Decision Tree, and Random Forest to construct different distance learning risk prediction models. We explored model portability between different disciplines. By comparing the Kappa value and F value of these models, it is found that the Random Forest Algorithm has the most stable performance. The predictive model constructed by the comprehensive curriculum was highly portable and could be interdisciplinary used. The learner's age, gender, education, daily homework performance, submission times, and the number of online question directly affect the learners' final academic performance.