Scholarlink

With the increasing prevalence of Parkinson’s disease (PD) with age, therefore, an early intervention is vital for better patient outcomes. The conventional methods in diagnostic analyses are mainly based on clinical examinations of motor symptoms and thus delayed until the disease has advanced significantly. This paper addresses the emergent challenge facing early treatment of Parkinson’s disease. The paper involves voice data analysis of the acquired data from UCI machine learning (ML) repository for early detection of PD. Our study analyses and compares four most widely used ML models for determining voice features predictive of PD, namely AdaBoost, Support Vector Machines, Logistic Regression, and Naïve Bayes. Furthermore, this paper involves robust data pre-processing techniques like missing value imputation, feature scaling, hyper-parameter tuning through Grid-searchCV, and five-fold cross validation techniques, also the utilization of SMOTE (Synthetic Minority Over-sampling Technique) in order to handle class imbalance in the acquired data. The results clearly show the superiority of AdaBoost model with high accuracy and precision of 97.44% and 96.97% respectively on the pre-processed training data, better than the other models compared. Therefore, aiming to open up huge opportunities for voice analysis in improving the predictive capability of ML models.

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

Enhanced Parkinson’s Disease Prediction Through Speech Data: A Machine Learning Approach

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Parambrata Sanyal,
  • Gopal Kumar Gupta,
  • Ajit K. Singh,
  • Bharti Kumari,
  • Praveen Kumar Gupta,
  • Akanksha Sahu

摘要

With the increasing prevalence of Parkinson’s disease (PD) with age, therefore, an early intervention is vital for better patient outcomes. The conventional methods in diagnostic analyses are mainly based on clinical examinations of motor symptoms and thus delayed until the disease has advanced significantly. This paper addresses the emergent challenge facing early treatment of Parkinson’s disease. The paper involves voice data analysis of the acquired data from UCI machine learning (ML) repository for early detection of PD. Our study analyses and compares four most widely used ML models for determining voice features predictive of PD, namely AdaBoost, Support Vector Machines, Logistic Regression, and Naïve Bayes. Furthermore, this paper involves robust data pre-processing techniques like missing value imputation, feature scaling, hyper-parameter tuning through Grid-searchCV, and five-fold cross validation techniques, also the utilization of SMOTE (Synthetic Minority Over-sampling Technique) in order to handle class imbalance in the acquired data. The results clearly show the superiority of AdaBoost model with high accuracy and precision of 97.44% and 96.97% respectively on the pre-processed training data, better than the other models compared. Therefore, aiming to open up huge opportunities for voice analysis in improving the predictive capability of ML models.