Improvement Of Rough Set Method For Cognitive Diagnosis

Cognitive diagnostic assessment(CDA)can not only evaluate the level of individual competence,but also provide the mastery of the attributes of each knowledge point,making targeted remedial teaching possible.Researchers have developed many parameterized cognitive diagnostic models and achieved good results,but there are also some problems,including the dependence of number of subjects and the long estimation time,above problems prevent cognitive diagnosis from being well applied to practical needs.In order to solve the problems,the researcher proposed to apply the rough set method based on classification algorithm to cognitive diagnosis.As a mathematical tool to describe and process uncertainty,the rough set can derive classification rules to classify objects in the universe without prior information.The cognitive diagnosis rough set method developed has the advantages: item parameters are not required;large number of subjects is also not required;suitable for tests with a plenty of attributes;do not need to equivalence;and can provide instant feedback of student diagnosis results.However,the existing cognitive diagnosis rough set method is still immature in attribute reduction method and decision rule judgment,and there are certain problems,such as the local convergence of the reduction result of genetic algorithm leads to insufficient reliability;The result of the method of judging the decision rules is rough;The performance cannot be improved under large samples;And hard to be used for the most common polytomous scoring items in current actual tests.In view of the above problems,this paper improves the two links of attribute reduction and decision rule judgment in the cognitive diagnosis rough set method,and extends the cognitive diagnosis rough set method to polytomous scoring tests.In order to explore the effect of the improved cognitive diagnosis rough set method and its influencing factors,this paper carried out four studies,including three simulation studies and one empirical study.In the first study,four representative attribute reduction methods of rough set are compared under different conditions under simulation conditions;in the second study,six machine learning methods are compared with the results of decision rules under the cognitive diagnostic rough set method;in the third study,by comparing the design of two kinds of decision tables,a polytomous scoring cognitive diagnosis rough set method was developed;the fourth study verifies the effectiveness of the improved cognitive diagnosis rough set method under empirical data.The results show that:(1)Under the four attribute hierarchical relations,the attribute reduction method of conditional information entropy performs better than other reduction methods in the rough set diagnosis method,including the genetic algorithm used in previous studies,and there is a significant difference in the accuracy between the method of conditional information entropy and the genetic algorithm.(2)As the number of attributes increases and the error rate of guess increases,the accuracy rate of DINA(deterministic inputs,noisy “and” gate)model decreases rapidly,while the rough set method is relatively more stable.(3)In most cases,among the six machine learning methods,support vector machines,random forests,and naive Bayesian classifiers perform better in rough set methods than simply using decision rules to judge the subject’s attribute mastery patterns,and the naive Bayesian classifier method performs best;When the number of attributes is large and the sample size is large,that is,when the training set is large for the machine learning method,the performance of the machine learning method is significantly better than the decision rule method,and there is a significant difference in the accuracy rate between the two methods.(4)Although the developed polytomous scoring cognitive diagnosis rough set method has a certain gap with the results of PC-DINA(partial credit DINA)model,its advantages of fast accuracy,independent of the number of subjects and good accuracy are still maintained,and it has certain practical value.(5)In the empirical study,the improved cognitive diagnosis rough set method is higher than DINA and the original cognitive diagnosis rough set method in the accuracy rate of pattern,and similar in the accuracy rate of attribute.