Comparison of decision trees used in data mining

Abstract

The purpose of this study is to compare decision trees obtained by data mining algorithms used in various areas in recent years according to different criteria. In the study, similar and different aspects of the decision trees obtained by different methods for classifying the students as successful and unsuccessful in terms of science literacy were revealed with the help of 12 independent variables included in the PISA 2015 student survey. Data collected across Turkey, from a total of 5895 students in the age group of 15, were analyzed in Java-based Weka software, which has an open source code. As a result of the analysis, it was found that the most successful algorithms in terms of correct classification rate were respectively Logistic Model, Hoeffding Tree, J.48, REPTree and Random Tree. In addition, regarding the decision trees obtained by different learning algorithms, variables that have been effective in the classification were found to be different. According to the results, it was concluded that independent variables found to be effective in the classification of the students for the decision trees obtained by different algorithms differed from each other and it was suggested to report the finding of more than one algorithm instead of those of only one algorithm.