Abstract:Parkinson's disease (PD), a neurological disorder renowned for its impact on hearing and movement, has become a focal point in recent scientific research. Timely and accurate detection of Parkinson's disease is crucial for effective intervention and treatment. Speech impairment serves as a common early symptom, emphasizing the significance of vocal disorders in PD diagnosis systems. Deep learning models show promise in accurately diagnosing Parkinson's disease, utilizing vocal signal processing and acoustic recognition techniques. However, a persistent challenge lies in selecting relevant acoustic features, crucial for improving accuracy and reducing computational complexity. This paper presents a novel Hybrid Ensemble Feature Selection (HEFS) Framework designed for a PD detection system based on a Deep Neural Network (DNN). The framework incorporates a multi-stage EFS process, starting with the utilization of three Feature Selection techniques—ReliefF, mRMR, and Chi-square—to individually extract relevant features from the dataset. Subsequently, Multiple Layers Dimensionality Reduction (MLDR) is applied to matrices derived from this HEFS. The MLDR process encompasses the normalization of matrix scores, score combinations, reconstruction of the new dataset, and, ultimately, feature reduction using the Neighborhood Component Analysis (NCA) algorithm. The proposed model in the study exhibits notable advancements in PD classification based on vocal features. The model achieved an impressive accuracy rate of 97.08% and an F1-score of 98.10%. Comparative analysis with state-of-the-art diagnostic methods reveals the superiority of the proposed model, surpassing recent techniques like CNN, Bi-LSTM, and Multi-Kernel SVM in accuracy. The model's AUC value of 0.98 further supports its excellence in classification performance compared to the original dataset (AUC = 0.90). Overall, the research significantly contributes to PD diagnosis by presenting a powerful approach that combines the HEFS-MLDR technique with a DNN model. This study successfully integrates the HEFS-MLDR technique with a DNN model, creating a robust and efficient framework. The innovative hybrid architecture, incorporating EFS technique and a dimensionality reduction concept, produces high-performing classification tools for PD. The achieved high accuracy and F1-score highlight the potential of this methodology for early and accurate PD detection.

Parkinson’s Disease Detection by Using Maximal Information Coefficient

A Recognition Method for Diabetic Retinopathy Based on Feature Selection

Distance Correlation Sure Independence Screening for Accelerated Feature Selection in Parkinson's Disease Vocal Data

Diagnosis of Parkinson's Disease with a Hybrid Feature Selection Algorithm Based on a Discrete Artificial Bee Colony

Enhancing Parkinson's Disease Prediction Using Machine Learning and Feature Selection Methods

Ensemble Machine Learning Approach for Parkinson's Disease Detection Using Speech Signals

Classification of Parkinson’s Disease Utilizing Multi-Edit Nearest-Neighbor and Ensemble Learning Algorithms with Speech Samples

A Multi-agent Feature Selection and Hybrid Classification Model for Parkinson's Disease Diagnosis

A Simple and Effective Approach Based on a Multi-Level Feature Selection for Automated Parkinson's Disease Detection

A Hybrid Feature Selection Algorithm Based on a Discrete Artificial Bee Colony for Parkinson's Diagnosis

Learning acoustic features to detect Parkinson's disease

HEFS-MLDR: A novel hybrid ensemble feature selection framework for improved deep neural network architecture in the diagnosis of Parkinson's disease

Machine learning approaches to identify Parkinson's disease using voice signal features

Vocal Feature Extraction-Based Artificial Intelligent Model for Parkinson's Disease Detection

Effective Dysphonia Detection Using Feature Dimension Reduction and Kernel Density Estimation for Patients with Parkinson's Disease.

Gene-related Parkinson's disease diagnosis via feature-based multi-branch octave convolution network

Detection of Parkinson disease using multiclass machine learning approach

FEATURE SELECTION USING EXTRA TREES CLASSIFIER FOR PARKINSON’S DISEASE CLASSIFICATION

Machine learning-based classification of Parkinson's disease using acoustic features: Insights from multilingual speech tasks

The prediction of Parkinson's disease based on Pearson coefficient feature screening and machine learning

Multi-Scale Feature and Multi-Channel Selection toward Parkinson's Disease Diagnosis with EEG