Scholarlink

Background <p>Prominence of Science, Technology, Engineering, and Mathematics (STEM) competencies has transcended mere academic achievements, contributing cross-cuttingly to individual, societal, and global development. Despite consensus that STEM-related outcomes are largely shaped by complex and interconnected factors, key determinants of STEM competencies remain insufficiently explored. Guided by Dynamic Model of Education Effectiveness (DMEE), the current study draws on Programme for International Student Assessment (PISA) 2022 dataset to identify key determinants of STEM competencies.</p> Methods <p>This study leverages data from 522,802 fifteen-year-old adolescents across 66 economies using six machine learning models, including Naïve Bayes, Decision Tree, Logistic Regression, Random Forest, AdaBoost, and XGBoost, to identify factors contributing to STEM competencies. In addition, this study examines model effectiveness by assessing predictive accuracy of each classifier.</p> Results <p>This study identified a key set of variables that predict STEM competencies, including home possessions, student part-time employment, gender, cognitive activation for reasoning, and mathematics anxiety. Model evaluation indicated that XGBoost classification outperformed other classifiers, with an average test accuracy of 78.3%.</p> Conclusions <p>Findings illuminate the complex interplay of student-, family-, and school-level factors in determining STEM competencies, suggesting the need to consider multifaceted, non-linear interactions across intra- and inter-personal levels, and design targeted interventions.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Predicting STEM competencies with machine learning: identifying key determinants among 522,802 adolescents

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月7日

查看全文
  • Ji Liu,
  • Dahman Tahri,
  • Millicent Aziku

摘要

Background

Prominence of Science, Technology, Engineering, and Mathematics (STEM) competencies has transcended mere academic achievements, contributing cross-cuttingly to individual, societal, and global development. Despite consensus that STEM-related outcomes are largely shaped by complex and interconnected factors, key determinants of STEM competencies remain insufficiently explored. Guided by Dynamic Model of Education Effectiveness (DMEE), the current study draws on Programme for International Student Assessment (PISA) 2022 dataset to identify key determinants of STEM competencies.

Methods

This study leverages data from 522,802 fifteen-year-old adolescents across 66 economies using six machine learning models, including Naïve Bayes, Decision Tree, Logistic Regression, Random Forest, AdaBoost, and XGBoost, to identify factors contributing to STEM competencies. In addition, this study examines model effectiveness by assessing predictive accuracy of each classifier.

Results

This study identified a key set of variables that predict STEM competencies, including home possessions, student part-time employment, gender, cognitive activation for reasoning, and mathematics anxiety. Model evaluation indicated that XGBoost classification outperformed other classifiers, with an average test accuracy of 78.3%.

Conclusions

Findings illuminate the complex interplay of student-, family-, and school-level factors in determining STEM competencies, suggesting the need to consider multifaceted, non-linear interactions across intra- and inter-personal levels, and design targeted interventions.