Scholarlink

<p>Cardiovascular disease (CVD) and diseases like diabetes present significant obstacles to public health, particularly in resource-constrained environments where timely diagnosis can profoundly influence patient outcomes. The process of selecting minimal influential features from the initial collection of features extracted from patients' diagnostics to classify healthy/diseased patients transforms the landscape. This experimental study is an attempt in the same direction of selection of the most influential minimal number of features by introducing an innovative approach for feature selection that culminates in a hybrid sampling optimization model (hLPUMA: Hybrid LASSO (Least Absolute Shrinkage and Selection Operator) and PUMA optimization algorithm). The proposed and implemented approach is validated using two distinct healthcare datasets: one focused on Indian Heart Disease (IHD) and the other on diabetes (DB). The essence of the approach lies in choosing the most insightful and diagnostic features that can greatly assist in the prompt and precise identification and categorization of illnesses in individuals. The proposed hLPUMA integrates the robust feature selection capabilities of LASSO with the efficient global search mechanism of PUMA, resulting in an improved method for identifying significant features. In the IHD dataset, hLPUMA identifies five significant clinical features, whereas in the DB dataset, it shortlists four essential features necessary for classifying individuals as healthy or infected. The enhanced feature groups are subsequently utilized to train an XGBoost classifier (employing a 70:30 train-test split and fivefold cross validation), resulting in impressive accuracy rates of 99.68% for IHD and 99.36% for the DB dataset, confirming that our approach is highly effective for medical diagnosis. To provide additional evidence of the effectiveness of the proposed model and its applicability across various contexts, we conduct an ablation study to evaluate the contribution of several selected features to performance. Furthermore, a Friedman test was conducted at the classifier level to obtain statistical validation of the appropriateness of XGBoost, deeming it the optimal model when paired with hLPUMA-selected features. We conducted a comparison of hLPUMA with seven well-established benchmarking optimization functions, including Sphere, Rastrigin, and Ackley, among others. The findings indicate that hLPUMA outperforms the conventional chi method, the LASSO model, the PUMA algorithm, and various hybrid combinations of these techniques when applied to these benchmark functions. This comprehensive investigation and its examination showcase the effectiveness of hybrid feature selection in creating patient-focused intelligent clinical decision support systems. The results hold enormous potential for incorporation into clinical decision support systems, facilitating swift, dependable, and comprehensible disease prediction, thereby assisting healthcare professionals in delivering prompt and effective treatment strategies.</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

A Novel Framework Comprising Fusion of Statistical Approach (LASSO) with Soft-Computing Algorithm (PUMA) for Optimal Feature Selection to Enhance Prediction of Cardiovascular Disease and Diabetes Disease

  • 期刊论文

  • 发表日期：

    2026年5月14日

查看全文
  • Amit Yadav,
  • Kamal Kant Hiran,
  • Prasun Chakrabarti,
  • Munish Khanna

摘要

Cardiovascular disease (CVD) and diseases like diabetes present significant obstacles to public health, particularly in resource-constrained environments where timely diagnosis can profoundly influence patient outcomes. The process of selecting minimal influential features from the initial collection of features extracted from patients' diagnostics to classify healthy/diseased patients transforms the landscape. This experimental study is an attempt in the same direction of selection of the most influential minimal number of features by introducing an innovative approach for feature selection that culminates in a hybrid sampling optimization model (hLPUMA: Hybrid LASSO (Least Absolute Shrinkage and Selection Operator) and PUMA optimization algorithm). The proposed and implemented approach is validated using two distinct healthcare datasets: one focused on Indian Heart Disease (IHD) and the other on diabetes (DB). The essence of the approach lies in choosing the most insightful and diagnostic features that can greatly assist in the prompt and precise identification and categorization of illnesses in individuals. The proposed hLPUMA integrates the robust feature selection capabilities of LASSO with the efficient global search mechanism of PUMA, resulting in an improved method for identifying significant features. In the IHD dataset, hLPUMA identifies five significant clinical features, whereas in the DB dataset, it shortlists four essential features necessary for classifying individuals as healthy or infected. The enhanced feature groups are subsequently utilized to train an XGBoost classifier (employing a 70:30 train-test split and fivefold cross validation), resulting in impressive accuracy rates of 99.68% for IHD and 99.36% for the DB dataset, confirming that our approach is highly effective for medical diagnosis. To provide additional evidence of the effectiveness of the proposed model and its applicability across various contexts, we conduct an ablation study to evaluate the contribution of several selected features to performance. Furthermore, a Friedman test was conducted at the classifier level to obtain statistical validation of the appropriateness of XGBoost, deeming it the optimal model when paired with hLPUMA-selected features. We conducted a comparison of hLPUMA with seven well-established benchmarking optimization functions, including Sphere, Rastrigin, and Ackley, among others. The findings indicate that hLPUMA outperforms the conventional chi method, the LASSO model, the PUMA algorithm, and various hybrid combinations of these techniques when applied to these benchmark functions. This comprehensive investigation and its examination showcase the effectiveness of hybrid feature selection in creating patient-focused intelligent clinical decision support systems. The results hold enormous potential for incorporation into clinical decision support systems, facilitating swift, dependable, and comprehensible disease prediction, thereby assisting healthcare professionals in delivering prompt and effective treatment strategies.