Scholarlink

<p>This study applies machine learning and probabilistic graphical modeling into the data from the PISA 2018 assessment for the purpose of examining how information and communication technology (ICT)-related factors and demographic characteristics simultaneously associate students’ mathematics achievement in Türkiye. After the stage of data cleaning and preprocessing, this analysis includes 6580 students from 186 schools. A random forest (RF) classifier is employed for predicting mathematics success, as it achieves an accuracy of 73% (SE = 0.01). With the aim of enhancing interpretability of this black-box model, permutation variable importance (PVI) and partial dependence (PD) plots are applied together with a Bayesian Network (BN) to reveal conditional dependencies among variables. The BN structure, which is learned through the Hill-Climbing algorithm and validated via 200 bootstrap replications, shows high structural and parametric stability. The achieved results indicate that “socio-economic status (ESCS)” and “school type (STRATUM)” have the strongest effect on mathematics achievement (MATHSUC). ICT-related factors such as “interest in ICT (INTICT)” and “leisure-time ICT use (ENTUSE)” are positively associated with MATHSUC, whereas “general ICT use at school (USESCH)” shows a negative relationship. The ICT-attitude subnetwork reveals a coherent developmental pattern in which students’ interest, competence, and social use of ICT mutually reinforce one another, reflecting a positive digital learning cycle. Overall, integration of RF and BN provides both predictive accuracy and interpretability, and offers a transparent and comprehensive framework for educational data analysis. These findings emphasize the collective importance of socio-economic context and digital engagement in promoting mathematics achievement and digital equity.</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

An Analysis of the Relationships Between Turkish Students’ Mathematics Achievement and ICT Variables: A Harmonization of Bayesian Network and Random Forest Methods

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月9日

查看全文
  • Ulku Hilal Yildirim,
  • Yasemin Kayhan Atilgan,
  • Derya Ersel

摘要

This study applies machine learning and probabilistic graphical modeling into the data from the PISA 2018 assessment for the purpose of examining how information and communication technology (ICT)-related factors and demographic characteristics simultaneously associate students’ mathematics achievement in Türkiye. After the stage of data cleaning and preprocessing, this analysis includes 6580 students from 186 schools. A random forest (RF) classifier is employed for predicting mathematics success, as it achieves an accuracy of 73% (SE = 0.01). With the aim of enhancing interpretability of this black-box model, permutation variable importance (PVI) and partial dependence (PD) plots are applied together with a Bayesian Network (BN) to reveal conditional dependencies among variables. The BN structure, which is learned through the Hill-Climbing algorithm and validated via 200 bootstrap replications, shows high structural and parametric stability. The achieved results indicate that “socio-economic status (ESCS)” and “school type (STRATUM)” have the strongest effect on mathematics achievement (MATHSUC). ICT-related factors such as “interest in ICT (INTICT)” and “leisure-time ICT use (ENTUSE)” are positively associated with MATHSUC, whereas “general ICT use at school (USESCH)” shows a negative relationship. The ICT-attitude subnetwork reveals a coherent developmental pattern in which students’ interest, competence, and social use of ICT mutually reinforce one another, reflecting a positive digital learning cycle. Overall, integration of RF and BN provides both predictive accuracy and interpretability, and offers a transparent and comprehensive framework for educational data analysis. These findings emphasize the collective importance of socio-economic context and digital engagement in promoting mathematics achievement and digital equity.