Towards a Prediction Model of Learning Performance: Informed by Learning Behavior Big Data Analytics

Using log data of 823 university students collected in two settings: their online learning setting and daily life setting (using campus ID cards for consumption purposes and book-borrowing in the university library), this study created indicators for online learning behavior, early-rising behavior, book-borrowing behavior and learning performance prediction. Five machine learning models were employed to analyze learning performance prediction, with the additional use of Boosting and Bagging to improve the accuracy of the prediction model. The predictability of the proposed model was also compared with that of both the Artificial Neural Network model and the Deep Neural Network model. At the same time, a classification rule set was established by combining decision tree and rule model, and a learning behavior diagnosis model combining decision tree and deep neural network was constructed. Findings showed that multi-scenario behavior performance indicators had strong predictive capabilities while the Deep Neural Network model had the highest prediction accuracy (82%) but was most time-consuming. The model based on the rule set is highly accurate, readable and operable and may be conducive to making accurate teaching interventions and resource recommendations.