Using Machine Learning to Predict Student Mathematics Performance in Six East Asian Countries: Evidence from PISA 2022

In an era increasingly defined by artificial intelligence and complex knowledge systems, robust mathematical performance is a cornerstone for the success of younger generations. While East Asian countries have consistently demonstrated exceptional results in international mathematics assessments, the intricate web of factors contributing to this achievement remains a subject of intense scholarly interest. A recent study, leveraging advanced machine learning techniques on data from the Programme for International Student Assessment (PISA) 2022, has shed new light on the most influential determinants of mathematics success among students in six high-performing East Asian nations and economies.

The research, published in Frontiers in Psychology, employed a sophisticated analytical pipeline to examine a comprehensive set of 62 variables affecting the mathematics achievement of nearly 27,000 fifteen-year-old students. The findings underscore the significant predictive power of machine learning models, with the XGBoost algorithm emerging as the most accurate, explaining approximately 57.03% of the variance in mathematics performance. This represents a notable leap in understanding compared to traditional statistical methods.

Key Findings: Self-Efficacy Reigns Supreme

At the forefront of the identified predictors is mathematics self-efficacy, a student’s belief in their ability to succeed in mathematics tasks. This factor exerted a substantially larger influence on performance than all other variables, highlighting the critical role of psychological confidence in academic achievement. Following closely were participation in extracurricular activities before school and the weekly hours dedicated to mathematics instruction. The study’s analysis revealed a consistent pattern: affective, behavioral, and instructional factors proved more predictive of mathematics success than structural and socioeconomic variables.

This emphasis on student-proximal determinants aligns with Self-Determination Theory (SDT), a framework positing that intrinsic motivation and optimal functioning are fostered by the fulfillment of three basic psychological needs: autonomy, competence, and relatedness. The study’s results suggest that within the educational contexts of East Asia, which often prioritize rigorous academic pursuit, a strong sense of mathematical competence (self-efficacy) is paramount for sustained engagement and high performance.

The East Asian Mathematics Phenomenon: A Deeper Dive

For decades, students from East Asian education systems, including Korea, Singapore, Japan, Chinese Taipei, Hong Kong, and Macao, have consistently topped international rankings in mathematics. This sustained excellence, often exceeding Western counterparts by the equivalent of one to two years of schooling, has fueled extensive research into the underlying cultural, pedagogical, and societal factors. These include a strong cultural emphasis on diligence and perseverance, high parental expectations, and a societal valorization of scholarly achievement. Pedagogically, these systems often feature coherent curriculum sequencing, systematic problem variation, and extensive practice.

However, this aggregate success masks significant internal variations. The PISA 2022 data allowed researchers to delve into the nuanced factors influencing performance within these elite systems, moving beyond broad generalizations. The study’s methodological rigor, utilizing advanced machine learning algorithms like Random Forest, LightGBM, XGBoost, AdaBoost, Elastic Net, and Linear Regression, allowed for the identification of the most influential predictors and the exploration of complex, non-linear relationships.

Unpacking the Predictors: Beyond Simple Correlations

The study’s detailed analysis, employing SHapley Additive exPlanations (SHAP) values, provided granular insights into how each predictor influences mathematics performance at an individual level.

• Mathematics Self-Efficacy: The overwhelming dominance of self-efficacy, as revealed by SHAP values, suggests that students who believe in their mathematical abilities are significantly more likely to achieve higher scores. This aligns with Bandura’s social cognitive theory and SDT, where perceived competence is a critical driver of motivation and effort. The study noted that this effect was largely gender-invariant, indicating its universal importance across genders within these contexts.
• Time Investment: While hours spent on mathematics per week and homework time were significant predictors, the analysis revealed diminishing returns and a degree of heterogeneity. This suggests that the quality of time spent on mathematics, rather than mere quantity, is more crucial. This finding has direct policy implications for countries where long study hours are common, hinting at the need to optimize instructional engagement over simply increasing time.
• Extracurricular Activities: Participation in extracurricular activities before school showed a complex, non-monotonic pattern. This complexity likely reflects the diverse nature of these activities, with academically focused ones potentially boosting performance while others might displace study time or lead to fatigue.
• Instructional Quality and Concept Familiarity: The positive impact of mathematics instruction quality and familiarity with mathematical concepts underscores the foundational importance of effective teaching and a solid grasp of core knowledge. This reinforces the value of investing in teacher professional development and curriculum design that emphasizes conceptual understanding.
• Educational Expectations: While generally positive, the effect of educational expectation levels was heterogeneous. This highlights the potential double-edged sword of high expectations in high-stakes environments, where aspirations must be aligned with intrinsic motivation to avoid anxiety and performance suppression.
• Digital Resources and Online Activities: The study noted a nuanced relationship with digital devices and online activities. While basic access to devices at home showed a modest positive correlation, extensive online activity did not automatically translate into higher performance, suggesting a need for mindful integration of digital tools to avoid displacing study time.
• Socioeconomic and Structural Factors: Consistent with much of the literature, socioeconomic status (SES) and home possessions showed a positive but modest influence. This suggests that in these high-achieving East Asian contexts, factors beyond material wealth, such as psychological and instructional elements, play a more dominant role in determining mathematics success.

Implications for Policy and Practice

The study’s findings carry significant implications for educators, policymakers, and researchers, particularly within the Confucian Heritage Culture (CHC) educational systems of East Asia.

• Prioritizing Self-Efficacy Development: Educational interventions should actively focus on building students’ confidence in their mathematical abilities. This can be achieved through mastery-oriented learning experiences, constructive feedback, and strategies to reduce mathematics anxiety.
• Optimizing Instructional Time: Instead of solely focusing on increasing study hours, educational systems should emphasize the quality of instruction. This includes fostering cognitively engaging, autonomy-supportive, and formatively assessed learning experiences.
• Promoting Equitable Learning Environments: The gender-invariant nature of key predictors like self-efficacy suggests that equitable outcomes can be achieved by ensuring that all students, regardless of gender, have access to need-satisfying learning conditions that foster competence, autonomy, and relatedness.
• Evidence-Informed Policy: The study’s use of machine learning and SHAP values provides a robust, data-driven foundation for policy decisions, moving beyond anecdotal evidence or traditional statistical analyses. This approach can help identify high-leverage intervention targets for maximum impact.

Methodological Advancements and Future Directions

This research not only provides valuable insights into mathematics achievement but also showcases the power of advanced machine learning techniques in educational research. The comprehensive preprocessing pipeline, the use of SHAP values for interpretability, and the rigorous model evaluation offer a replicable framework for future studies.

However, the study acknowledges its limitations, including its cross-sectional design, which precludes definitive causal claims, and reliance on self-reported data. Future research should incorporate longitudinal data, multi-source assessments (including teacher and administrative data), and more diverse international samples to further validate and expand upon these findings. Investigating the interplay between student-level factors and school- and teacher-level influences through multilevel modeling will also be crucial for a more holistic understanding.

In conclusion, this study offers a compelling argument for the integration of machine learning in understanding complex educational phenomena. By identifying mathematics self-efficacy as a paramount predictor and highlighting the nuanced roles of time, instruction, and motivation, the research provides actionable insights for enhancing mathematics education, not only in East Asia but potentially across global systems striving for academic excellence and equity.