By applying educational data mining methods to big data related to large-scale exams, functional relationships are discovered in a basic sense and hidden pattern(s) can be revealed. Within the scope of the research, to show how the self-organizing map (SOM) method can be used in terms of educational data mining, how SOM differs from other clustering methods in terms of visual outputs (map) and how to interpret the outputs, and it is aimed to give information about how effective the variables are in grouping individuals into groups according to the answers given to the items. In this study, students of OECD countries participating in the 2015 PISA were modeled using SOM and the outputs of the created model were examined. In this respect, the study can be accepted as a descriptive survey model. According to the results of the analysis, outputs were obtained for the educational process of the data set, the state of neurons, neighborhood distance, code vectors, heat maps, the number of clusters and the distribution of the number of students to countries and clusters. At the same time, it was determined that 4 clusters were formed according to the analysis results, and the most effective variables in clustering by examining the heat maps were perceived feedback from science teachers, teacher-directed science instruction, average of plausible values in science, enquiry based science instruction and adaptive instruction in science lessons. Researchers who want to clearly determine the effectiveness of the input variables in cluster analysis can be advised to use SOM.