Scholarlink

<p>Early and reliable diagnosis of Parkinson’s disease (PD) is a critical concern in biomedical diagnostics, mainly owing to the high dimensionality and heterogeneity among multimodal datasets. Early and accurate diagnosis of PD constitutes a major challenge in biomedicine diagnostics, given the highly dimensional and heterogeneous nature of multimodal datasets. In this paper, proposed Linear Discriminant Analysis Optimization Machine Learning (LDA-OML) model estimate the features with min–max probabilistic optimization for the estimation of variables in the images. The feature selection is performed with the Optimized Flamingo-integrated Dolphin Algorithm (OFDA). LDA extracts discriminative features to maximize class separability, while min–max probabilistic optimization enhances feature discriminability and mitigates intermodal redundancy. The OFDA algorithm adaptively identifies a compact subset of high-impact features, further refined through a ranking-based learning mechanism to optimize classification performance. The simulation analysis is performed for the datasets, the UCI Parkinson Disease Spiral Drawings, capturing dynamic kinematic trajectories, and the Parkinson’s Drawings Dataset on Kaggle, which includes static spiral and wave pictures. Experimental evaluation demonstrates that LDA-OML achieves superior accuracy (up to 97.25%), significant feature reduction, and reduced computational overhead, exhibiting significant performance over existing approaches. OFDA, as the feature selection technique, the framework selects the most important features to improve classification efficiency. With the proposed LDA-OML on datasets from UCI and Kaggle, classification accuracy was 98.3%, precision was 97.5%, recall was 98.7, and the AUC-ROC score achieved by the proposed method was 99.2%. For comparison, methods such as SVM (accuracy 94.2%), RF (accuracy 92.5%) and XGB (accuracy 96.8%) improved performance in terms of separability and scalability through the epochs, where the proposed LDA-OML framework rendered considerable performance enhancements and achieved the maximum accuracy of 97.4% at 200 epochs. The proposed method offers a robust, interpretable, and efficient solution for early PD detection, with broad applicability in medical imaging, digital biomarkers, and behavioural analysis.</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

LDA-OML: A Novel Framework for Enhanced Parkinson’s Disease Diagnosis

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月7日

查看全文
  • Poonam Yadav,
  • Meenu Vijarania,
  • Preeti Sharma,
  • Supreet Singh,
  • Madhu Shukla,
  • A. Johnson Santhosh

摘要

Early and reliable diagnosis of Parkinson’s disease (PD) is a critical concern in biomedical diagnostics, mainly owing to the high dimensionality and heterogeneity among multimodal datasets. Early and accurate diagnosis of PD constitutes a major challenge in biomedicine diagnostics, given the highly dimensional and heterogeneous nature of multimodal datasets. In this paper, proposed Linear Discriminant Analysis Optimization Machine Learning (LDA-OML) model estimate the features with min–max probabilistic optimization for the estimation of variables in the images. The feature selection is performed with the Optimized Flamingo-integrated Dolphin Algorithm (OFDA). LDA extracts discriminative features to maximize class separability, while min–max probabilistic optimization enhances feature discriminability and mitigates intermodal redundancy. The OFDA algorithm adaptively identifies a compact subset of high-impact features, further refined through a ranking-based learning mechanism to optimize classification performance. The simulation analysis is performed for the datasets, the UCI Parkinson Disease Spiral Drawings, capturing dynamic kinematic trajectories, and the Parkinson’s Drawings Dataset on Kaggle, which includes static spiral and wave pictures. Experimental evaluation demonstrates that LDA-OML achieves superior accuracy (up to 97.25%), significant feature reduction, and reduced computational overhead, exhibiting significant performance over existing approaches. OFDA, as the feature selection technique, the framework selects the most important features to improve classification efficiency. With the proposed LDA-OML on datasets from UCI and Kaggle, classification accuracy was 98.3%, precision was 97.5%, recall was 98.7, and the AUC-ROC score achieved by the proposed method was 99.2%. For comparison, methods such as SVM (accuracy 94.2%), RF (accuracy 92.5%) and XGB (accuracy 96.8%) improved performance in terms of separability and scalability through the epochs, where the proposed LDA-OML framework rendered considerable performance enhancements and achieved the maximum accuracy of 97.4% at 200 epochs. The proposed method offers a robust, interpretable, and efficient solution for early PD detection, with broad applicability in medical imaging, digital biomarkers, and behavioural analysis.