| Cognitive diagnostic assessment(CDA)can not only assess the individual ability value level,but also provide the mastery of the attributes of each knowledge point,making targeted remedial teaching possible.Diagnostic classification model with polytomous attribute examines the subjects which level of attributes have mastered by detailed items in more detail.which is more compliant with "Full-time compulsory education mathematics curriculum standard(experimental manuscript)" about Multilevel assessment requirements for different knowledge and skills.However,the commonly used parameter estimation methods in cognitive diagnosis(such as EM algorithm,MCMC)requires a large sample size to achieve stable and reliable estimation results,while those kind of samples do not meet the actual teaching situation,so It is not suitable for using in real world teaching evaluation.Support Vector Machine(SVM)can obtain more accurate classification in a small sample size,which is very suitable for practical teaching.So this article applies the SVM model to diagnostic classification model with polytomous attribute in a small sample to provide a convenient and effective tool for actual teaching evaluation.In order to explore the effect of the SVM model on diagnostic classification model with polytomous attribute and the influencing factors of the diagnosis effect,four studies in this article: Study 1 mainly compares the classification accuracy of pG-DINA model and SVM model in simulated data;Study 2 and Study 3 explore the factors that affect the classification accuracy of the SVM model in the application of a diagnostic classification model with polytomous attribute in simulated data.Study 4 compared the classification accuracy of pG-DINA model and SVM model in empirical data.The research indicates:(1)When the attribute has no attribute hierarchy,for the accuracy rate,pG-DINA is higher in the simulated data while SVM is higher in the empirical data.When the attribute has an attribute hierarchy,the accuracy rate of SVM is higher.(2)When the attribute has an attribute hierarchy,the accuracy of child attributes in the SVM model is not completely affected by the parent attributes.(3)For attribute multi-level cognitive diagnosis,when the SVM model performs classification,it is the best to choose the RBF kernel function,and the training set size is about 100.(4)In small sample,as the number of attributes and the maximum number of attributes increasing,the accuracy of the SVM model will be decreased.It is recommended that the number of attributes should not exceed 5,the maximum number of attributes should not exceed 3 in small sample test,and it is not recommended to use the suggested extreme value for both at the same time.(5)It is recommended to use the SVM model for classification.In the actual evaluation with small sample size. |
