Scholarlink

<p>Parkinson's disease is a chronic neurodegenerative condition that affects motor function and frequently involves significant changes in voice and speech. Early and consistent detection of parkinson's disease can greatly enhance patient outcomes by enabling timely intervention. This study has performed an exhaustive comparative study of five popular machine learning models–Logistic Regression, Multilayer Perceptron (MLP), Support Vector Machine, Random Forest, and Gradient Boosting each model was tested with 2–4 optimization techniques, such as Adam, L-BFGS, and LIBLINEAR. Using Parkinson's dataset from the UCI Machine Learning Repository, the experiment has used repeated stratified K-fold cross-validation to obtain performance estimates. This study has assessed each model–optimization technique pair using accuracy, precision, recall, and F1-score metrics and has used Friedman and Wilcoxon tests to check if differences are statistically significant. The results indicate that MLP with Adam achieves the best cross-validation accuracy (91.35%), followed by Random Forest with gini criterion (91.04%). Gradient Boosting with exponential loss reaches ~ 88.77%. Logistic Regression and Support Vector Classifier (SVC) resulted in lower accuracies (~ 83.02% and 85.58%). These results highlight the importance of the role that optimizers and hyperparameter tuning play in classification performance for the detection of parkinson's disease.</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

Comparative analysis of optimization algorithms in machine learning models for Parkinson's disease prediction

  • 期刊论文

  • 发表日期：

    2025年11月17日

查看全文
  • Aditya Singhal,
  • Akash Soni,
  • Shweta Meena

摘要

Parkinson's disease is a chronic neurodegenerative condition that affects motor function and frequently involves significant changes in voice and speech. Early and consistent detection of parkinson's disease can greatly enhance patient outcomes by enabling timely intervention. This study has performed an exhaustive comparative study of five popular machine learning models–Logistic Regression, Multilayer Perceptron (MLP), Support Vector Machine, Random Forest, and Gradient Boosting each model was tested with 2–4 optimization techniques, such as Adam, L-BFGS, and LIBLINEAR. Using Parkinson's dataset from the UCI Machine Learning Repository, the experiment has used repeated stratified K-fold cross-validation to obtain performance estimates. This study has assessed each model–optimization technique pair using accuracy, precision, recall, and F1-score metrics and has used Friedman and Wilcoxon tests to check if differences are statistically significant. The results indicate that MLP with Adam achieves the best cross-validation accuracy (91.35%), followed by Random Forest with gini criterion (91.04%). Gradient Boosting with exponential loss reaches ~ 88.77%. Logistic Regression and Support Vector Classifier (SVC) resulted in lower accuracies (~ 83.02% and 85.58%). These results highlight the importance of the role that optimizers and hyperparameter tuning play in classification performance for the detection of parkinson's disease.