Scholarlink

Background <p>This study develops a leakage-free neutrosophic regression model to identify robust determinants of academic success and benchmarks its predictive value against a conventional logistic model.</p> Methods <p>Data from 865 first-year students were analysed via T/I/F transformations derived from standardized scores and weighted aggregation (w<sub>T</sub> = 0.5, w<sub>I</sub> = 0.3, w<sub>F</sub> = 0.2). Leakage-prone grade variables were removed. Performance was assessed via 10-fold cross-validation, repeated CV, bootstrap optimism, calibration curves, Brier scores, SHAP, permutation importance, and decision curve analysis. A multilevel model was used to evaluate school and region clustering. A complete sensitivity analysis assessed the predictive stability across 150 weight combinations on the (w<sub>T</sub>, w<sub>I</sub>, w<sub>F</sub>) simplex.</p> Results <p>The neutrosophic model demonstrated intense discrimination (AUC = 0.757; 95% CI: 0.702–0.812), stable cross-validated performance (AUC = 0.708 ± 0.060), and superior calibration (Brier = 0.189). Sensitivity analysis revealed consistently high AUC values (0.74–0.76) across weight configurations, with optimal performance occurring for w<sub>T</sub> ≈ 0.5–0.7, confirming robustness to T/I/F weighting choices. The key predictors included WASSCE mathematics, BECE mock scores, parental occupation, and years spent at home. Decision curve analysis revealed superior net benefits for the neutrosophic model at low-to-moderate thresholds (0.05–0.40). Multilevel modelling indicated negligible school or regional clustering (ICC ≈ 0). Forecast-style simulations further revealed that early-stage assessments produce the most significant changes in prediction success.</p> Conclusion <p>Neutrosophic regression provides an uncertainty-aware, interpretable framework with strong calibration and stable performance, offering a valuable tool for early risk identification and educational decision-making in Ghana’s teacher-training system.</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

Modelling academic success under uncertainty in Ghanaian colleges of education using neutrosophic regression

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月27日

查看全文
  • Senyefia Bosson-Amedenu,
  • Francis Ayiah-Mensah,
  • Anthony Joe Turkson,
  • John Awuah Addor,
  • Emmanuel Mensah Baah

摘要

Background

This study develops a leakage-free neutrosophic regression model to identify robust determinants of academic success and benchmarks its predictive value against a conventional logistic model.

Methods

Data from 865 first-year students were analysed via T/I/F transformations derived from standardized scores and weighted aggregation (wT = 0.5, wI = 0.3, wF = 0.2). Leakage-prone grade variables were removed. Performance was assessed via 10-fold cross-validation, repeated CV, bootstrap optimism, calibration curves, Brier scores, SHAP, permutation importance, and decision curve analysis. A multilevel model was used to evaluate school and region clustering. A complete sensitivity analysis assessed the predictive stability across 150 weight combinations on the (wT, wI, wF) simplex.

Results

The neutrosophic model demonstrated intense discrimination (AUC = 0.757; 95% CI: 0.702–0.812), stable cross-validated performance (AUC = 0.708 ± 0.060), and superior calibration (Brier = 0.189). Sensitivity analysis revealed consistently high AUC values (0.74–0.76) across weight configurations, with optimal performance occurring for wT ≈ 0.5–0.7, confirming robustness to T/I/F weighting choices. The key predictors included WASSCE mathematics, BECE mock scores, parental occupation, and years spent at home. Decision curve analysis revealed superior net benefits for the neutrosophic model at low-to-moderate thresholds (0.05–0.40). Multilevel modelling indicated negligible school or regional clustering (ICC ≈ 0). Forecast-style simulations further revealed that early-stage assessments produce the most significant changes in prediction success.

Conclusion

Neutrosophic regression provides an uncertainty-aware, interpretable framework with strong calibration and stable performance, offering a valuable tool for early risk identification and educational decision-making in Ghana’s teacher-training system.