Scholarlink

<p>Chronic kidney disease (CKD) is a prevalent and serious global health issue, with a significant impact on individuals globally. Hence, it is imperative to promptly obtain an accurate diagnosis and interpretation for the commencement of appropriate treatment as timely detection and intervention can enhance the probability of long-term survival. Existing projection-based methods for feature selection do not yield desired outcomes due to their different objectives necessitating the need for innovation approaches for a higher predictive performance. This study proposes a novel fusion-based feature selection (FFS) model for the optimization and selection of distinct features to enhance CKD diagnosis. This study utilizes the University of California, Irvine (UCI) CKD dataset and addresses missing data and imbalance issues through Multiple Imputations by Chain Equation (MICE) and Borderline Synthetic Minority Oversampling Technique (Borderline-SMOTE). The proposed model integrates different machine learning (ML) classifiers, conventionally known as black boxes, with SHAP values to provide interpretability and gain transparency in the decision-making process. The proposed FFS model performs better than single feature selection approaches, achieving 100% in all of the evaluation metrics for support vector machine, light gradient boosting, random forest, voting and extreme gradient boosting classifiers compared to other existing literature that also utilized the same dataset. Notably, the SHAP analysis reveals that features such as red blood cell, white blood cell count and the pus cell clumps show model specific interactions. This aids healthcare in understanding and effectively applying the model’s outputs. Empirical evidence demonstrates that our proposed approach exhibits superior performance which has the potential to complement physicians’ diagnosis of kidney diseases. Also, the incorporation of explainability enhances the clarity of outcomes and facilitates the identification of the underlying cause of the diseases, contributing to more transparency and ethically sound AI applications in healthcare.</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

FFS-IML: fusion-based statistical feature selection for machine learning-driven interpretability of chronic kidney disease

  • 期刊论文

  • 发表日期：

    2025年4月21日

查看全文
  • Grace Ugochi Nneji,
  • Happy Nkanta Monday,
  • Venkat Subramanyam Reddy Pathapati,
  • Saifun Nahar,
  • Goodness Temofe Mgbejime,
  • Edwin Sunday Umana,
  • Md Altab Hossin

摘要

Chronic kidney disease (CKD) is a prevalent and serious global health issue, with a significant impact on individuals globally. Hence, it is imperative to promptly obtain an accurate diagnosis and interpretation for the commencement of appropriate treatment as timely detection and intervention can enhance the probability of long-term survival. Existing projection-based methods for feature selection do not yield desired outcomes due to their different objectives necessitating the need for innovation approaches for a higher predictive performance. This study proposes a novel fusion-based feature selection (FFS) model for the optimization and selection of distinct features to enhance CKD diagnosis. This study utilizes the University of California, Irvine (UCI) CKD dataset and addresses missing data and imbalance issues through Multiple Imputations by Chain Equation (MICE) and Borderline Synthetic Minority Oversampling Technique (Borderline-SMOTE). The proposed model integrates different machine learning (ML) classifiers, conventionally known as black boxes, with SHAP values to provide interpretability and gain transparency in the decision-making process. The proposed FFS model performs better than single feature selection approaches, achieving 100% in all of the evaluation metrics for support vector machine, light gradient boosting, random forest, voting and extreme gradient boosting classifiers compared to other existing literature that also utilized the same dataset. Notably, the SHAP analysis reveals that features such as red blood cell, white blood cell count and the pus cell clumps show model specific interactions. This aids healthcare in understanding and effectively applying the model’s outputs. Empirical evidence demonstrates that our proposed approach exhibits superior performance which has the potential to complement physicians’ diagnosis of kidney diseases. Also, the incorporation of explainability enhances the clarity of outcomes and facilitates the identification of the underlying cause of the diseases, contributing to more transparency and ethically sound AI applications in healthcare.