Research On The Construction Of The Model Of Predicting Academic Failure Risk And Early Warning Feedback Design In Distance Education

With the popularization of distance education,the monitoring and improvement of distance education quality has become the focus.There are many distance learners,but there are only a few teachers or their teaching teams,and the teacher-student ratio is very low.And because of the teaching method of time and space separation,it is more and more important to study how to reduce the risk of learners' learning failure,stimulate learning motivation,and help teachers achieve personalized teaching in the case of time and space separation.Effective prediction and timely feedback are important guarantees for improving teaching and learning.By summarizing previous studies on the construction of the model to predict learners' learning success or failure and provide feedback,it is found that the previous research scenes are focused on MOOC,Blended learning,etc.And there are problems such as a single research course,a small number of research samples,poor portability of the constructed prediction model,and insufficient effectiveness of early warning intervention feedback.Based on this,this study collected background information and online learning behavior data of 14,448 distance learners from four different disciplines of a network and continuing education college in Chongqing as a research sample.This research uses data mining algorithms to construct Distance Learning Risk Prediction Models,and uses User-Centered Design Theory and Participatory Design Methods to design feedback.The main research contents include:First,this study uses data mining technology to construct a distance learning risk prediction model,and studies whether it is possible to build an interdisciplinary Distance Learning Risk Prediction Models based on educational big data.Among them,in the data preprocessing stage,this study selects two algorithms of Smote Sampling and Costsensitive,so as to solve the problem of uneven distribution of data samples of learner risk categories.And compare it with the effect of the constructed distance learning risk prediction model.Second,this study uses Chi-Square Test,Correlation Analysis and Multiple Linear Regression Analysis to analyze the correlation between learner background information and online learning behavior characteristics and final academic performance in different disciplines.And the analysis results are combined with the distance education teaching environment to explain and put forward suggestions that are helpful to distance education teaching.Third,this study uses User-Centered Design Theory and Participatory Design Methods to design early warning feedback.A set of participatory design tools applied in the field of learning analysis are proposed to obtain user expectations and needs.Including focus group,persona profiles,knowledge mapping,learner journey,feedback prototype construction.Finally,the designed early warning feedback prototype uses questionnaires and interviews for usability evaluation.Research indicates:(1)The comprehensive course Distance Learning Risk Prediction Models constructed by the random forest algorithm is highly portable.Its Kappa value is 0.5863 and F value is 0.782,which can be used across disciplines.Using Smote sampling to deal with the uneven distribution of the sample of the learner risk category data can greatly improve the performance of the trained prediction model.(2)The performance of predictive features in different disciplines is not exactly the same.Synthetic features(sex*education)help predictive model training and analysis of predictive features.Distance learners' age,usual homework grades and gender*qualifications are important characteristics that affect academic performance in different disciplines.The total number of homework submissions is an important feature that affects academic performance only in liberal arts courses.The total number of online learning and the total number of online questions are only important features that affect academic performance in science courses.The results of predictive feature analysis provide guidance for the design of early warning feedback.Although distance learning risk prediction models can be used across disciplines,the design of early warning feedback needs to consider different disciplines separately.(3)Most of the users prefer to participate in the early warning feedback designed in the quasi-experimental study.According to the data of the analysis questionnaire,learners are very satisfied with the design of the early warning feedback prototype in the four dimensions of perceived usefulness,cognitive load,user satisfaction and self-direction,and the overall evaluation satisfaction is high.According to the analysis of the data of the interview content,the personalized report on learning warnings greatly meets the warning needs of teachers and administrators of learning centers.It is mainly reflected in the low cognitive load of the feedback interface,the reasonable design of the early warning section,the rich feedback information and the comprehensive data indicators.This provides effective evidence support for remote learning early warning intervention.