Deep Learning-Based Method for Detecting Parkinson using 1D Convolutional Neural Networks and Improved Jellyfish Algorithms

Abstract

Parkinson's disease (PD) is a common disease that predominantly impacts the motor scheme of the neural central scheme. While the primary symptoms of Parkinson's disease overlap with those of other conditions, an accurate diagnosis typically relies on extensive neurological, psychiatric, and physical examinations. Consequently, numerous autonomous diagnostic assistance systems, based on machine learning (ML) methodologies, have emerged to assist in evaluating patients with PD. This work proposes a novel deep learning-based classification of Parkinson's disease (PD) using voice recordings of people into normal, idiopathic Parkinson, and familial Parkinson. The improved jellyfish algorithm (IJFA) is utilized for hyper-parameter selection (HPS) of a 1D convolutional neural network (1D-CNN). The proposed technique makes use of the significant elements of 1D-CNN and filter-based feature selection models. Because of their strong performance in dealing with noisy data, the filter-based algorithms Relief, mRMR, and Fisher Score were chosen as the top choices. Using just 62 characteristics, the combination of deep relief features and deep learning was able to discriminate between people. The competence of the proposed 1D-CNN with IJFA method was determined through specific network metrics. The proposed 1D-CNN with IJFA method attains a total accuracy of 98.6%, which is comparatively better than the existing techniques. The proposed model produced around 9.5% improvements in accuracy, respectively, when compared to the data obtained without dimensionality reduction.