An Approximation of Label Distribution-Based Ensemble Learning Method for Online Educational Prediction

Online education becomes increasingly important since traditional learning is shocked heavily by COVID-19. To better develop personalized learning plans for students, it is necessary to build a model that can automatically evaluate students’ performance in online education. For this purpose, in this study we propose an ensemble learning method named light gradient boosting channel attention network (LGBCAN), which is based on label distribution estimation. First, the light gradient boosting machine (LightGBM) is used to predict the performance in online learning tasks. Then The Channel Attention Network (CAN) model further improves the function of LightGBM by focusing on better results in the K-fold CrossEntropy of LightGBM. The results are converted into predicted classes through post-processing methods named approximation of label distribution to complete the classification task. The experiments are employed on two datasets, data science bowl (DSB) and answer correctness prediction (ACP). The experimental results in both datasets suggest that our model has better robustness and generalization ability