Scholarlink

<p>Parkinson’s disease (PD) presents a neurological challenge affecting individuals with and without motor skills, particularly manifesting as speech difficulties during its initial phases. The successful diagnosis of PD often relies on the detection of these speech problems. Numerous initiatives are still being implemented to solve these problems and improve the accuracy of the diagnosis of PD. However, several computational techniques have been effective in identifying PD in its early stages, bypassing the drawbacks of conventional diagnostic techniques. This study proposed an ensemble approach that combines machine learning (ML) and convolutional neural network (CNN) techniques to detect Parkinson’s disease (PD) in its early stages. For our investigation, we used the well-known PD dataset from the University of California, Irvine (UCI) repository. The proposed ensemble approach is developed using two base methods: the multilayer perceptron (MLP) and a hybrid model called ParkRLP. To create the new dataset, the ParkRLP and MLP networks are level-0 learners, while the lightweight PD-CNN model is a level-1 learner. Successful PD detection is achieved by the proposed ensemble strategy using the level-1 classifier as the PD-CNN model. The lightweight PD-CNN model consists of eight layers. The conv-1D, Maxpooling-1D, and Dense layers have been used to modify the PD-CNN model. According to the dependent and independent factors in the newly generated dataset, the meta-classifiers were utilized for training. Then, the classifiers’ outputs are determined after they have been trained and tested at level-0. The efficacy of the proposed model was further underscored by its consistent 5% improvement over previous research on individual ML and CNN models. In conclusion, out of all the techniques for detecting PD, the proposed ensemble method of the lightweight CNN approach had the highest accuracy, at 99.47%, and an F1-score of 99.50%. Experimental findings demonstrate the accuracy of the proposed methodology in estimating PD and its simplicity for implementation in medical settings for diagnostic purposes.</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 lightweight CNN-based ensemble approach for early detecting Parkinson’s disease with enhanced features

  • 期刊论文

  • 发表日期：

    2025年4月7日

查看全文
  • Dip Kumar Saha,
  • Tushar Deb Nath

摘要

Parkinson’s disease (PD) presents a neurological challenge affecting individuals with and without motor skills, particularly manifesting as speech difficulties during its initial phases. The successful diagnosis of PD often relies on the detection of these speech problems. Numerous initiatives are still being implemented to solve these problems and improve the accuracy of the diagnosis of PD. However, several computational techniques have been effective in identifying PD in its early stages, bypassing the drawbacks of conventional diagnostic techniques. This study proposed an ensemble approach that combines machine learning (ML) and convolutional neural network (CNN) techniques to detect Parkinson’s disease (PD) in its early stages. For our investigation, we used the well-known PD dataset from the University of California, Irvine (UCI) repository. The proposed ensemble approach is developed using two base methods: the multilayer perceptron (MLP) and a hybrid model called ParkRLP. To create the new dataset, the ParkRLP and MLP networks are level-0 learners, while the lightweight PD-CNN model is a level-1 learner. Successful PD detection is achieved by the proposed ensemble strategy using the level-1 classifier as the PD-CNN model. The lightweight PD-CNN model consists of eight layers. The conv-1D, Maxpooling-1D, and Dense layers have been used to modify the PD-CNN model. According to the dependent and independent factors in the newly generated dataset, the meta-classifiers were utilized for training. Then, the classifiers’ outputs are determined after they have been trained and tested at level-0. The efficacy of the proposed model was further underscored by its consistent 5% improvement over previous research on individual ML and CNN models. In conclusion, out of all the techniques for detecting PD, the proposed ensemble method of the lightweight CNN approach had the highest accuracy, at 99.47%, and an F1-score of 99.50%. Experimental findings demonstrate the accuracy of the proposed methodology in estimating PD and its simplicity for implementation in medical settings for diagnostic purposes.