Design And Application Of Online Learning Achievement Prediction Framework Based On Multi-algorithm Fusion

Early prediction of academic performance is a key element in solving the problems of high dropout rate and low participation in online learning,and achieving online learning performance prediction is an urgent scientific issue to be solved.How to extract influential factors with high predictive power and construct a high-performance learning achievement prediction framework are two important steps in learning achievement prediction.In terms of selecting factors that affect academic performance,existing studies have mostly relied on the results of one or two methods of selecting factors that affect academic performance to determine the final predictive factors.However,the predictive factors extracted from research have their own advantages,and the predictive factors and methods are not very transferable.In terms of prediction methods,existing studies mainly use a single algorithm to train a single classifier and an ensemble classifier to predict learning outcomes.However,a single algorithm has problems such as low prediction accuracy and poor transferability.Therefore,it is necessary to analyze the performance and applicability of the selection methods for factors affecting academic performance,in order to provide the necessary dominant features for constructing a framework for predicting academic performance;Apply the idea of multi algorithm fusion to the field of education,design an online learning performance prediction framework based on multi algorithm fusion,and explore its prediction accuracy and application effect in practical teaching environments.This paper collected the data of an educational technology class of a university in northeast China learning the course of Web Design and Development on Moodle learning platform in 2021 academic year.After data cleaning and other preprocessing steps,we used six methods to select factors affecting learning performance,including Pearson coefficient,to build a prediction dataset.Based on the above dataset,we used six algorithms,including Bayesian network,to build a single classifier,On the basis of comparing the predictive performance of various classifiers,choose the best method for selecting factors that affect learning performance.Combining the stacking fusion framework,this article designs an online learning performance prediction framework based on multi-algorithm fusion.After comparing the performance of six algorithms such as Naive Bayes,a better performance prediction algorithm is selected to train the first layer classifier,and a logistic regression algorithm is used to combine the first layer classifier to predict learners’ online learning performance categories.To further verify the reliability and transferability of the multi algorithm fusion prediction framework in practical teaching,the study applied online learning data from another group of learners on the Moodle platform for analysis.The purpose of predicting academic performance is not only to obtain the predicted results,but more importantly,to identify learners’ learning problems and provide learning improvement plans based on the predicted results.This article is based on the learning behavior logic of "early warning feedback-self adjustment-academic performance improvement",and conducts a4-week teaching early warning experiment for high-risk and medium risk learners from the9 th to 12 th weeks of the course.After the course is over,a comparative analysis is conducted before the end of the course The next 8 weeks of learning data will be used to test the effectiveness.The data results show that each classifier has good predictive performance on the dataset extracted by the Pearson correlation coefficient method.In this paper,the information gain rate method is selected to extract factors affecting learning performance.The predictive performance of the online learning performance prediction framework based on multi-algorithm fusion improved the predictive performance of the single classifier trained by a single algorithm and the ensemble classifier,respectively,with a prediction accuracy of78.9%.In the validation dataset,the optimal prediction results were still achieved through multi-algorithm fusion,indicating that the online learning performance prediction framework based on multi-algorithm fusion can better capture the relationship between learning features and grades.In the early warning experiment,there was a significant difference in the learning data of high and medium risk category learners from the first and last 8 weeks,and their academic performance significantly improved.This indicates that the visualization of warning content such as online learning performance prediction results in actual teaching helps learners realize potential academic risks,actively participate in learning activities,significantly improve the quality of online course learning,and ensure the resilience and persistence of online learning.