Veri Madenciliği Yöntemleri İle Ülkelerin Pisa Başarı Düzeylerini Etkileyen Değişkenlerin İncelenmesi

Abstract

The aim of this research is to determine the important variables that predict the PISA 2018 reading field achievement score of countries with different achievement levels, using 34 independent variables obtained from the student questionnaire given to the students who participated in PISA in 2018. For this purpose, 79 countries that entered PISA were ranked according to their success percentages and according to this ranking, these countries were divided into lower, middle and upper group countries.Then, three countries were selected from each of the lower group, middle group and upper group countries and nine different samples were obtained, all of these nine countries were determined as the study sample. The sample of Turkey, which is one of the three countries selected for the middle group, was also included in the study independently. Then, data mining analyzes were carried out on Turkey, the sample of lower, middle, upper group countries and the study sample using logistic regression, Classification and Regression Tree and Random Forest methods. It has been observed that the number of important variables that predict reading comprehension success can be reduced from 34 to a number between two and eight. Like this; Data mining classification prediction models, which can predict the PISA success level, were obtained by using a small number of variables. It was determined that the models obtained had high predictive performance in the two-category (unsuccessful-successful) prediction of success and acceptable in the three-category (low, medium-high) prediction. According to the results obtained, among the 34 independent variables, first of all, PISA test's perception of difficulty, reading pleasure, father's education level, perception of reading difficulty, socio-economic level index, meaning of life, teacher's direction of education and weekly test language learning time were used in different estimation models of 27 variables were found to be important variables.