Research And Application Of Hybrid Learning Behavior Analysis And Learning Early Warning Algorith

With the improvement of education informatization,the scale of blended teaching combining online and offline is expanding.Using the learning behavior data recorded on the platform,students’ real learning status can be analyzed,and students with learning difficulties can be found in the early stage.However,in the current research,feature selection mainly relies on domain experts,and model construction ignores the problems of category imbalance and sparseness in learning behavior data and pays too much attention to the model itself.In view of the above problems,this paper uses SHAP model to analyze and select features,constructs an early warning model of academic performance classification based on SMOTE-XGBoost-FM,and develops a learning early warning system based on the early warning model.The specific work is as follows :Firstly,an interpretable learning behavior feature selection method based on SHAP model is proposed.This method uses SHAP value to evaluate the influence of features on the prediction results,calculates the contribution of each feature to all possible feature combinations,obtains the SHAP value of each feature according to the calculation results,and sorts the features according to the absolute value of SHAP value to select the best feature for prediction.Finally,compared with the traditional feature selection method,the experimental results show that the feature selected according to the SHAP value has better generalization ability for prediction.In addition,the SHAP model-based interpretable method can analyze individual student samples and provide reference for teachers to analyze students’ learning status.Secondly,a blended academic performance early warning model based on SMOTE-XGBoost-FM is proposed.First,the sample class imbalance problem in the data set is solved by SMOTE sampling.Aiming at the problem that the model cannot capture the correlation between data due to data sparseness,XGBoost is used to train the sampled data set for the first time,and the features are crossed.The leaf nodes where the samples are trained are encoded to generate high-order features.The generated high-order feature data is combined with the original features to form a new data set as the input of the factorization machine(FM).Finally,iterative training is performed to obtain the optimal model.The experimental results show that the accuracy of SMOTE-XGBoost-FM in blended learning achievement warning reaches 92.7 %,which is 5.7 % and 11.7 % higher than that of single XGBoost and FM models,respectively.Finally,a learning achievement early warning system is designed and implemented.The back-end of the system adopts the Django framework.The system is mainly divided into teacher-side,student-side and administrator-side.The analysis results and prediction results of the data are visually displayed through Boostrap and Layui technologies.The functional and non-functional tests of the system are carried out.The results show that the system designed in this paper meets the expectations.The system facilitates teachers to grasp the learning dynamics of learners in time,discover the learning risks of learners in advance,and intervene in a targeted manner.