Predicting mathematics achievement using process data: A comparison of linear regression and random forest

This study aimed to construct a predictive model of mathematics achievement using computer-based assessment data from the mathematics domain of PISA 2022 in Korea. The model incorporated process data derived from item characteristics (cognitive level, item type, discrimination, and difficulty) along with learner background variables. Six models were developed: model 1 included only background variables; models 2 to 5 included background variables and process data based on a single item characteristic; and model 6 combined all variables. Linear regression and random forest algorithms were applied to compare predictive performance and identify key predictors. The results showed that random forest outperformed linear regression overall, and models 2 to 5 were selected as final models. The most important predictors included process data, mathematics self-efficacy, and socioeconomic and cultural status, demonstrating the effectiveness of process data in predicting mathematics achievement.