Prediction Of Students' Academic Level Based On Feature Selection And Stacking Framework

Online learning has become a trend in people's learning life,and the prediction of online learning students' the academic level is an inevitable problem at present and in the future.It can predict students' academic level accurately according to their daily performance and some personal characteristics,which will play a great role in students' understanding of self-learning degree,and can provide guiding significance in targeted teaching and academic early warning.In order to explore the role of data quality and fusion model in predicting students' academic performance,and to improve the prediction ability of the model,this paper constructed a prediction model of students' academic performance based on feature selection and Stacking framework,and named it the Stacking fusion model.The specific work results and research results of this paper are as follows:Firstly,according to the diversity of data features,including discrete features and continuous features,this paper analyzes and processes the meaning,type,balance and missing value of data features.In view of the fact that the ID3 model cannot process the continuous data,this paper discretizes the continuous data by using the method of equal width division to ensure the consistency of the input data of all the tree models in the experiment.According to the requirements of sklearn library on data,all data were numerically mapped to ensure the operability of the experiment.Secondly,in view of the blindness and uncertainty of the selection of the base model in the Stacking frame fusion process,the paper added the process of model selection before the prediction model construction to select the model that can be used for fusion and the model that can be used for feature selection.In this paper,ID3 algorithm,CART algorithm and random forest algorithm are used to construct the prediction model of students' academic level respectively.According to the prediction results of the model,the model is evaluated and selected.Through model selection,the two models that are most suitable for fusion are directly selected and used for fusion experiment,which avoids multiple fusion experiments and saves operation time.Through model selection,the best performance model is selected for feature selection,which can retain effective features to the greatest extent and avoid data loss.Thirdly,this paper constructs a prediction model of students' academic performance based on feature selection and Stacking framework.Firstly,the tree model with the best performance among the three models(random forest prediction model)is used to select the characteristics of the data samples,so as to filter the redundant data and ensure the effectiveness of the input data.In addition,based on the two-layer Stacking framework,XGBoost algorithm was used to fuse the first and second tree models(random forest prediction model,ID3 prediction model)in the three models to construct the Stacking fusion model,thus improving the model accuracy and classification accuracy in each category.By combining feature selection with Stacking framework,the model accuracy and generalization ability were greatly improved.Through parameter optimization,Stacking fusion model is proposed in this paper.The final model prediction accuracy reached 90.63%,the L category classification accuracy is 93%,the M category classification accuracy is 90%,and the H type classification accuracy is 88%.Compared with a single decision tree model and random forest model,the accuracy and classification accuracy are greatly improved.In addition,by analyzing the importance of characteristics of Stacking fusion model,this paper puts forward corresponding suggestions for academic warning and targeted teaching.