Predicting reading achievement and mathematics achievement with PISA 2022 student survey data

Artificial neural networks are computer systems that mimic the learning function of the human brain and perform the learning process with examples. These networks consist of interconnected processing units consisting of artificial nerve cells, and each connection has a weight value. The network's information is stored in these weight values and spread throughout the network. Unlike traditional computing methods, artificial neural networks are adaptive systems that can work with incomplete information and make decisions under uncertainty. Multilayer perceptron networks are the most widely used model of artificial neural networks today. These networks can provide valid results, especially in solving engineering problems. The PISA survey is organized by the OECD every three years and measures the basic knowledge and skills of 15-year-old students. Turkey has been participating in PISA since 2003, and the focus in 2022 is on mathematical literacy. According to PISA 2022 data, the relationship between students' socioeconomic status, family education status, parental attitudes and academic achievement is examined through various survey questions and answers. The aim of the study is to measure the extent to which the differences between the socioeconomic status and parental attitudes mentioned above can predict student success, and also to compare the prediction success between the Rapid Miner deep learning network and the Python multilayer artificial deep learning network. For this purpose, the SPSS format data published on the OECD website on December 5, 2023 were downloaded, the student surveys applied to students and the parent surveys applied to parents were examined, and variables containing labels related to economic issues and labels related to parental attitudes were selected. The mathematical literacy and reading skill scores that constitute the dependent variables of the study were published as 10 reasonable reading skill scores and 10 reasonable mathematical literacy scores. First, the average of the 10 published scores was taken and then coded between 0-6 as specified in the Ministry of National Education PISA 2024 report. In addition, in order to see to what extent the selected independent variables measure student success, the averaged mathematical skill and reading skill scores were coded as (0-unsuccessful) and (1-successful) according to the average scores of the OECD countries. The selected independent variables, dependent variables coded between (0-6) and (0-1), were transferred to the artificial neural network created in the Rapid Miner program and Python programming language to predict the scores of mathematics skills and reading skills, and the rate of independent variables predicting the dependent variables was determined. As a result of the analysis, the success rate of economic variables in predicting mathematics skills and reading skills scores was found to be 39%-43%, and the success rate of family education status and parental attitudes variables in predicting mathematics skills and reading skills scores was found to be 38%-42%. The rate of economic variables predicting students' mathematics skills and reading skills success was found to be 77%-78%, and the rate of family education status and parental attitudes variables predicting mathematics skills and reading skills success was found to be 78%-77%. In this context, it was found that economic elements and parental attitudes were important predictors of student success. In addition, the deep learning network created in the Rapid Miner program made 10% more successful predictions compared to the deep learning network created with the Python program and Keras library.