| “China’s Educational Modernization 2035” is a strategic policy document for the continued development of education informationization in our country.It emphasizes the importance of individualized teaching and collaborative learning,particularly encouraging the use of big data to conduct personalized analysis of students,to promote the organic combination of large-scale education and personalized training,and to promote intelligent education throughout the entire process of teaching,learning,practice,and evaluation.In addition,with the rapid development of global education big data and artificial intelligence technology,datadriven teaching decision-making models are gradually becoming an important measure to promote educational and teaching reforms.Therefore,AI-enabled intelligent education is not only necessary for implementing the Party’s education policies,but also necessary for the development of “weak artificial intelligence” to “strong artificial intelligence”.Currently,academic performance is still an important indicator for measuring the academic performance of college students,which not only relates to whether students can graduate successfully,but also affects the overall quality of student training in universities.In this context,course grade prediction is a key task,and its accuracy directly determines whether student problems can be identified in a timely manner and their learning status adjusted.Currently,grade warning mainly has two directions: one is to use students’ historical courses and other educational information to predict the course grades of the new semester,and to issue warnings based on the difference between the predicted grade value and the passing score;the other is to directly predict whether students have risks or their risk levels,and issue warnings based on classification results.However,difficulties in data collection,poor data quality,and single algorithm models often result in less-than-ideal grade prediction models.Therefore,improving the accuracy of grade prediction models and broadening the information output width of models remain important challenges.To address the aforementioned issues,this thesis introduces the idea of conditional density estimation to construct a course grade distribution prediction model.This model utilizes incomplete information to predict the complete distribution of grades.The main research content of this thesis is as follows:(1)Prediction and Warning of Grades Distribution Based on Conditional Density EstimationTo address the issue of low accuracy in predicting grades due to limited information and complex factors affecting students’ performance,this thesis proposes a new method based on probability density estimation.The method introduces a mixture density network to design a grade distribution prediction model,which can directly output the complete grades distribution for the target course.To ensure consistency between the output probabilities and the actual probabilities,we use a calibration algorithm to calibrate the model,further improving the accuracy and reliability of the prediction results.Finally,we calculate the probability of a student’s course grade falling within the failing range based on the grade distribution.Compared with traditional methods,this research method has higher prediction accuracy and more decision-making information,providing strong support for student academic warning.(2)“Smart Data” Academic Comprehensive Analysis Intelligent Early Warning SystemTo promote the advancement of intelligent technology in education governance,we have designed a “Smart Data” achievement analysis and comprehensive warning system that can quickly integrate and apply models.This system aggregates data from various sources in the education field through co-construction and sharing,and uses machine learning algorithms to timely and effectively present students’ academic risks to educators,helping them better understand the deeper meaning of the data,identify problems in student learning,and provide precise intervention measures. |
