| Identifying abnormalities in psychological and educational testing is critical to uncovering flaws in test design or abnormal behavior of subjects and the different response behaviors of the subjects showed significant differences in item response and response time patterns.In this dissertation,rapid guessing behavior recognition in the test are studied by combining the response and response time information.Aiming at the rapid guessing of the subjects,the item response mixture model was constructed by combining the response and response time.Using a Bayesian approach to estimate parameters in the model.In this dissertation,the Pólya-Gamma(PG)Gibbs sampling combination with the MCMC(PG-MCMC)algorithm is used to realize the Bayesian estimation of the model based on the Logistic and to improve the efficiency of parameter estimation.Simulation studies show that compared with MCMC algorithm,PG-MCMC can perform parameter estimation more efficiently,with lower Bias,RMSE and SD,has higher true detection rate(TDR)and lower false detection rate(FDR)for rapid guessing.Missing item observations is common in psychological and educational testing.In this dissertation,a suitable missing mechanism model is constructed for the ignorable missing of observed data,and the XGBoost method is used to identify the subjects who take rapid guessing behavior in the test.Simulation studies show that XGBoost can more efficiently and accurately identify subjects who take rapid guessing behavior in tests;when the observed data is not missing,considering the response and response time at the same time,the detection effect of rapid guessing behavior is better;Compared with the parameter estimation method,XGBoost can guarantee the small class in the case of unbalanced data categories has a high TDR;When there are ignorable missings in the observed data,XGBoost Sparsity-aware Split Finding algorithm can accurately classify subjects who take different behaviors without preprocessing of missing imputation. |
