Development of machine learning based computerized adaptive testing application for Minnesota multiphasic personality inventory

This study highlights adaptive tests in a computer environment that offered individual differences. MMPI, which has 13 factors and 566 items and an important place in personality tests, has been studied in this context. Data were collected from 970 people over the age of 18 to decrease the number and duration of items and make the test more efficient. In CAT performed with machine learning, entropy values for item selection, entropy discordance error for the cutpoint at test termination, Bayesian Classifier was used for classification. In addition, Parallel Analysis was used to determine the unidimensionality, and KR-20 analyzes were used for reliability. To determine the classification accuracy, the 95% case belonging to the posterior probability class in the leave one out a method in the training-test sets for machine learning was determined as the best. Entropy, a measure of disorder in a given data set was calculated and the highest entropy value was given to the participant as the first item. To evaluate the practicality of the CAT, the test lengths of the paper-pencil application and the CAT were compared, and it was found that it decreased between 85%-92% in all subscales. The classifications are at a reasonable level with an error rate of 0,05, and the highest classification value was reached in the L subscale. The limitations of the classic measurement tool in terms of practicality can be accomplished with the CAT, which allows more efficiency by producing comparable results.