Scholarlink

<p>Using routinely collected clinical data for the early detection of liver disease is often challenging owing to class imbalance, limited feature dimensionality, and nonlinear relationships among biochemical markers. This study evaluates hybrid machine learning models of liver disease classification on a set of 583 patient records with ten clinical-biochemical features (Indian Liver Patient Dataset, ILPD). We performed standard preprocessing such as missing-value handling, feature scaling, and categorical encoding. Logistic regression baselines, support vector machines, shallow neural networks, and hybrid ensemble models based on stacking were tested. Repeated stratified 10 × 5 cross-validation was used to guarantee reliable estimation given a limited sample size, and the results were reported with mean performance, variability, and 95% confidence intervals. Hybrid models, in particular stacking ensembles, are found to have consistently higher balanced accuracy and ROC–AUC than individual classifiers and are more robust across validation folds. The work is framed as a methodological benchmarking study rather than a clinically deployable diagnostic 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

Hybrid machine learning framework for early detection of liver disease using SVM, neural networks, and stacking models

  • 期刊论文

  • 发表日期：

    2026年6月15日

查看全文
  • M. S. Sreenivasa Rao,
  • Chintalapudi V. Suresh,
  • V. Nageshwar,
  • K. Nagaraju

摘要

Using routinely collected clinical data for the early detection of liver disease is often challenging owing to class imbalance, limited feature dimensionality, and nonlinear relationships among biochemical markers. This study evaluates hybrid machine learning models of liver disease classification on a set of 583 patient records with ten clinical-biochemical features (Indian Liver Patient Dataset, ILPD). We performed standard preprocessing such as missing-value handling, feature scaling, and categorical encoding. Logistic regression baselines, support vector machines, shallow neural networks, and hybrid ensemble models based on stacking were tested. Repeated stratified 10 × 5 cross-validation was used to guarantee reliable estimation given a limited sample size, and the results were reported with mean performance, variability, and 95% confidence intervals. Hybrid models, in particular stacking ensembles, are found to have consistently higher balanced accuracy and ROC–AUC than individual classifiers and are more robust across validation folds. The work is framed as a methodological benchmarking study rather than a clinically deployable diagnostic system.