Özet:

In this study, multi category classification was performed by using R (R Core Team, 2013) programming language and simulating an individualised classification test on computer. Two category, unidimensional 500 items and 1000 person data were created in R programme. According to the number of classification categories, the average classification accuracy (ACA) and average test length (ATL) values for test effectiveness, correlation (r) between actual and predicted abilities for measurement accuracy, bias, root mean square error (RMSE), mean absolute error (MAE) values were determined with the average of 25 repetitions. In the study, 16 conditions were created with 2 ability estimation methods as Weighted likelihood estimation (WLE), Expected posterior distribution (EPD); 4 item selection methods based on Maximum fisher information (MFI), Kullback leibler information (KLB) ability based (EB) and Cut score based (KN), 2 classification criteria as Confidence interval (CI) and Sequential Probability Ratio Test (SPRT). In three and four category classification, the Maximum fisher information estimated ability based item selection method gave better results in terms of test length, classification accuracy with weighted likelihood ability estimation. Expected posterior distribution ability estimation with Kullback leibler cut score based item selection method made the most accurate classification with the least number of items in two category classification. Expected posterior distribution ability estimation showed the best performance for two category classification when crossed with the confidence interval classification criterion.

Anahtar Kelimeler: