Research On Online Test Prediction Based On LightGBM Model

With the development of Internet technology and the change of education philosophy,as well as the impact of the COVID-19 on traditional classroom teaching,more and more students are learning and acquiring knowledge through online platforms.This change in education methods has put forward higher requirements for the intelligence and Personalization of online education platforms.The intelligent education platform should be able to accurately predict students’ current learning effects,assess students’ knowledge mastery,and determine students’ knowledge weaknesses,so as to provide targeted guidance and recommend the most appropriate teaching resources.This thesis takes the historical learning records of Santa intelligent education system in the past 3 years as the research object,takes LightGBM model as the prototype,carries out feature extraction as well as discovery through feature interaction,feature statistical information,learning behavior feature perspective,dynamically constructs ELO scoring grading method for students and questions,and Introducing the forgetting mechanism of knowledge and other researches,combining SHAP feature selection and embedded feature selection,the ELO-LightGBM model is proposed,and the quantified system of student learning effect is established,which improves the generalization ability of LightGBM model and enhances the prediction effect of the model.Subsequently,the ELO-LightGBM model was applied to the Riiid Answer Correctness Prediction competition on the Kaggle data science competition platform,and obtained a score of 0.7928(AUC metric)with one quarter dataset training,which improved 1.49% compared to the organizer’s benchmark score and achieved better prediction results.At the same time,the simulation experiment proves that the ELO-LightGBM model also has a greater prediction effect compared with machine learning models such as SVM,logistic regression and random forest,which illustrates the advantages of the model and has important implications for the intelligence and personalization of online education platforms.