Scholarlink

Background <p>Elevated low-density lipoprotein cholesterol (LDL-C) is a major risk factor for cardiovascular disease, yet conventional clinical thresholds often fail to capture complex interactions among demographic, metabolic, and behavioral factors. Machine-learning (ML) approaches offer improved classification and risk stratification beyond traditional cutoffs, particularly when combined with explainable methods.</p> Methods <p>We analyzed WHO STEPS survey data (2018 onward) from low- and lower-middle-income countries, including 61,780 adults (44,782 normal and 16,998 abnormal cholesterol). Missing values were imputed using K-nearest neighbors, and categorical variables were numerically encoded. Recursive Feature Elimination with cross-validation (RFECV) using Random Forest identified the top 15 predictors. Nine ML classifiers were trained on an 80/20 train–test split and evaluated using accuracy, precision, recall, F1-score, and area under the ROC curve (AUC). Model interpretability was assessed using permutation feature importance implemented through ELI5.</p> Results <p>Recursive feature elimination feature selection method highlighted metabolic and socio-demographic factors including alcohol consumption, fasting blood sugar, age group, weight, education level, marital status, and residency as the most influential predictors of abnormal cholesterol. Ensemble models consistently outperformed traditional classifiers. Random Forest achieved the best performance (accuracy = 96.66%, F1-score = 96.58%, AUC = 0.993), demonstrating strong discriminatory ability across precision–recall thresholds. Permutation analysis further identified hypertension status, alcohol use, physical activity, waist category, and residency as key determinants.</p> Conclusion <p>Explainable ensemble-based ML models effectively capture the multifactorial determinants of cholesterol abnormality in population-level data. This transparent approach offers a robust and practical tool for early cardiovascular risk stratification in resource-limited settings.</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

Transparent machine learning-based classification of cholesterol levels using random forest and ELI5 explainability

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月15日

查看全文
  • Fetlework Gubena Arage,
  • Tsegasilassie Gebremariam Abate,
  • Tigist Kifle Tsegaw,
  • Eyob Akalewold Alemu,
  • Yeelshaday Asrat Tekle,
  • Maregu Biresaw Eyasu,
  • Zinabu Bekele Tadese

摘要

Background

Elevated low-density lipoprotein cholesterol (LDL-C) is a major risk factor for cardiovascular disease, yet conventional clinical thresholds often fail to capture complex interactions among demographic, metabolic, and behavioral factors. Machine-learning (ML) approaches offer improved classification and risk stratification beyond traditional cutoffs, particularly when combined with explainable methods.

Methods

We analyzed WHO STEPS survey data (2018 onward) from low- and lower-middle-income countries, including 61,780 adults (44,782 normal and 16,998 abnormal cholesterol). Missing values were imputed using K-nearest neighbors, and categorical variables were numerically encoded. Recursive Feature Elimination with cross-validation (RFECV) using Random Forest identified the top 15 predictors. Nine ML classifiers were trained on an 80/20 train–test split and evaluated using accuracy, precision, recall, F1-score, and area under the ROC curve (AUC). Model interpretability was assessed using permutation feature importance implemented through ELI5.

Results

Recursive feature elimination feature selection method highlighted metabolic and socio-demographic factors including alcohol consumption, fasting blood sugar, age group, weight, education level, marital status, and residency as the most influential predictors of abnormal cholesterol. Ensemble models consistently outperformed traditional classifiers. Random Forest achieved the best performance (accuracy = 96.66%, F1-score = 96.58%, AUC = 0.993), demonstrating strong discriminatory ability across precision–recall thresholds. Permutation analysis further identified hypertension status, alcohol use, physical activity, waist category, and residency as key determinants.

Conclusion

Explainable ensemble-based ML models effectively capture the multifactorial determinants of cholesterol abnormality in population-level data. This transparent approach offers a robust and practical tool for early cardiovascular risk stratification in resource-limited settings.