Predictive role of demographics, maths anxiety, and enjoyment in pupils’ standardised maths achievement: a machine learning approach

This study seeks to determine the factors affecting the achievements of Grade 5 primary school pupils in the Italian national assessment, INVALSI, using Machine Learning (ML) regression algorithms. Four ML algorithms were applied to data from 25,827 Grade 5 Italian primary school pupils who participated in the 2017–2018 INVALSI mathematics test. The Boosting model achieved the highest explanatory power (R2 = .152), with mathematics anxiety (relative importance 40.2), ESCS (37.0), and enjoyment (17.0) emerging as the dominant predictors. Random Forest (R2 = .142) and Decision Tree (R2 = .097) models confirmed this pattern, consistently ranking anxiety (46.1%), ESCS (33.5%), and enjoyment (14.0%) as the three most important factors. Regularised Linear Regression (R2 = .127) revealed directional effects: ESCS (β = .213) and enjoyment (β = .130) positively predicted achievement, while anxiety (β = −.180) was negatively associated. Across all four algorithms, gender and macroregion demonstrated minimal predictive importance (≤ 6.4% and ≤ 2.3%, respectively).