Scholarlink

Background <p>Intestinal parasitic infections remain a persistent public health concern in low-resource settings, including Somaliland. Despite being preventable, deworming coverage among preschool-aged children remains low. This study aimed to estimate the prevalence of deworming uptake, identify predictive factors associated with deworming uptake, and compare the performance of multiple machine learning algorithms using the Somaliland Demographic and Health Survey (SDHS 2020).</p> Methods <p>A total of 2,753 children aged 1–5 years were included. Descriptive and inferential analyses were conducted to estimate deworming prevalence and examine associated factors. The dataset was preprocessed and split into 80% training and 20% testing subsets. Eight machine learning algorithms—Random Forest (RF), XGBoost (XGB), LightGBM (LGB), K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Logistic Regression (LR), Gaussian Naive Bayes (GNB), and Decision Tree (DT)—were trained to predict deworming uptake. Class imbalance was addressed using the Synthetic Minority Over-sampling Technique (SMOTE). Model performance was evaluated using accuracy, area under the receiver operating characteristic curve (AUC), precision, recall, specificity, and F1 score. Feature importance was assessed using Boruta and SHAP interpretability methods.</p> Results <p>The prevalence of deworming among preschool-aged children in Somaliland was 6.8%. Deworming uptake varied across sociodemographic groups, with differences observed by residence type, region, household wealth, maternal age, place of delivery, maternal education, and partner’s employment status. Among the evaluated models, the K-Nearest Neighbors algorithm demonstrated the most balanced predictive performance (AUC: 0.827; F1-score: 0.27), indicating improved ability to identify the minority class of dewormed children. Although ensemble tree-based models such as Random Forest showed higher precision, SHAP analysis based on the Random Forest model identified type of residence, wealth index, partner’s employment, and place of delivery as the most influential predictive features.</p> Conclusion <p>Deworming coverage among preschool-aged children in Somaliland remains substantially below international targets and is characterized by marked socioeconomic and geographic disparities. Machine learning models demonstrated good predictive performance in identifying population subgroups with lower likelihood of deworming uptake. These findings provide evidence to support data-informed strategies aimed at improving equitable access to preventive chemotherapy services.</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

Machine learning prediction of deworming uptake among preschool aged children in Somaliland using 2020 Somaliland demographic and health survey data

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月17日

查看全文
  • Hamse Arab Ali,
  • Asma Khadar,
  • Abdulkadir Mohamed Nuh,
  • Hamse Adam Abdi,
  • Osman Mohamed Osman,
  • Abdisalam Hassan Muse,
  • Mahomoud Sulaiman Muse

摘要

Background

Intestinal parasitic infections remain a persistent public health concern in low-resource settings, including Somaliland. Despite being preventable, deworming coverage among preschool-aged children remains low. This study aimed to estimate the prevalence of deworming uptake, identify predictive factors associated with deworming uptake, and compare the performance of multiple machine learning algorithms using the Somaliland Demographic and Health Survey (SDHS 2020).

Methods

A total of 2,753 children aged 1–5 years were included. Descriptive and inferential analyses were conducted to estimate deworming prevalence and examine associated factors. The dataset was preprocessed and split into 80% training and 20% testing subsets. Eight machine learning algorithms—Random Forest (RF), XGBoost (XGB), LightGBM (LGB), K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Logistic Regression (LR), Gaussian Naive Bayes (GNB), and Decision Tree (DT)—were trained to predict deworming uptake. Class imbalance was addressed using the Synthetic Minority Over-sampling Technique (SMOTE). Model performance was evaluated using accuracy, area under the receiver operating characteristic curve (AUC), precision, recall, specificity, and F1 score. Feature importance was assessed using Boruta and SHAP interpretability methods.

Results

The prevalence of deworming among preschool-aged children in Somaliland was 6.8%. Deworming uptake varied across sociodemographic groups, with differences observed by residence type, region, household wealth, maternal age, place of delivery, maternal education, and partner’s employment status. Among the evaluated models, the K-Nearest Neighbors algorithm demonstrated the most balanced predictive performance (AUC: 0.827; F1-score: 0.27), indicating improved ability to identify the minority class of dewormed children. Although ensemble tree-based models such as Random Forest showed higher precision, SHAP analysis based on the Random Forest model identified type of residence, wealth index, partner’s employment, and place of delivery as the most influential predictive features.

Conclusion

Deworming coverage among preschool-aged children in Somaliland remains substantially below international targets and is characterized by marked socioeconomic and geographic disparities. Machine learning models demonstrated good predictive performance in identifying population subgroups with lower likelihood of deworming uptake. These findings provide evidence to support data-informed strategies aimed at improving equitable access to preventive chemotherapy services.