A Machine Learning Cognitive Diagnosis Method With Tolerable Missing Values

In psychological and educational evaluation,missing data is a very important research question.There are some common missing value processing methods,such as EM,EI and so on.However,the lack of data in cognitive diagnosis cannot satisfy the hypothesis of random data.Because the missing date in cognitive diagnosis usually has different kinds of reasons.The most important thing is that personalized diagnosis should be provided for students for cognitive diagnosis,instead of using group values to fill in the data like the traditional missing data processing.At present,there are few researches on the imputation of missing data in cognitive diagnosis,and there is no specific filling method for missing data of cognitive diagnosis.Whether these traditional filling methods can be directly applied to cognitive diagnosis data or not still needs more researches.Presently,the existing cognitive diagnosis methods diagnose complete data by default,including parametric one and nonparametric one.Among the machine learning algorithms,there is a kind of algorithm that can tolerate the existence of missing values.This study hopes to introduce this kind of machine learning algorithm into cognitive diagnosis,and develop it into a cognitive diagnosis method which does not rely on interpolation,has tolerance for missing values and has a high accuracy rate.Through three studies,the effectiveness of this method will be verified.Research 1 mainly constructs the method and compares it with the existing methods to investigate its characteristics.Research 2 verifies the characteristics of Xgboost diagnostic method for missing data processing.Research 3 is to verify its effectiveness through empirical data.The specific conclusions are as follows:(1)Xgboost cognitive diagnosis method Xgboost performs better than other methods when the number of topics is small and the sliding probability is high.(2)Xgboost can get better accuracy without processing missing data.(3)When the missing ratio is low,the missing data processing method has little influence on the results of Xgboost cognitive diagnosis method.