Data Mining Model for Forecasting Academic Performance of Undergraduates Based on Behavioral Features and SVM in E-learning

E-learning plays an increasingly important role in modern education. Behavioral data from online education platforms make it possible to mine college students' academic performance. The existing data mining models pay little attention to the weight of features in academic performance prediction, and most of them simply treat all features equally. However, each characteristic has a different influence on the predicted outcome. Taking account of this fact, we proposed a data mining model for forecasting undergraduates' academic performance based on multi-feature and support vector machine (SVM). First, the original data was extracted from the e-learning platform and processed. Then, each feature was measured its correlationship with academic performance according to Pearson’s rule and given a weight. Finally, the data mining model was constructed based on the weighted muti-feature and SVM. Simulation of MATLAB shows that the proposed method has higher Accuracy and Recall on forecasting academic performance of undergraduates.