An exploration of factors affecting PISA performance using CHAID analysis and C4.5 algorithm

The purpose of this research is to identify the factors that most influence students' mathematical literacy in the PISA 2018 samples of Turkey, Peru, and China using the C4.5 Algorithm and CHAID Analysis, and to compare the performance of these two algorithms on the same datasets. The conducted analyses were performed after removing all missing data from the datasets, and they were applied with 5,959 student data from the Turkey, 2,241 student data from the Peru, and 12,058 student data from the China. The dependent variable, which is mathematical literacy, was transformed into a categorical variable, categorized as below average and above average. The independent variables consisted of 22 items related to self-perception, adaptability, sense of belonging, family support, and school environment concepts. After organizing the dataset, two decision tree methods were separately applied and reported for each of the three samples. The analysis results indicated that the factors that most influence mathematical literacy were self-perception and family support in the Turkey sample, adaptability in the Peru sample, and family support and school environment in the China sample. According to the analysis results, different outcomes were obtained in the studies conducted on the two samples using the C4.5 and CHAID algorithms. It was observed that the C4.5 algorithm had a lower error rate and produced more accurate classification nodes in the Turkey and Peru samples. However, in the China sample, the decision tree generated by the CHAID analysis contained more accurate classification nodes. Nevertheless, it was determined that the C4.5 algorithm was more suitable for the scale structure of the classification options used in all countries.