Neuron degeneration in the human brain causes Parkinson's disease (PD). The technique employed for detecting early PD is based on a parametric analysis of acoustic sounds. Machine learning methods such as SVM, KNN, decision tree, and random forest are used in a significant manner. When compared to machine learning, deep-learning (DL) algorithms always produce greater results. DL algorithms are utilized to assess the patient's voice record. In this work, the authors introduce the MAYFLY optimization algorithm, a nature-inspired metaheuristic optimization technique. The approach utilizes convolutional neural networks (CNNs) coupled with bi-directional gated recurrent units (BiGRUs) to leverage the temporal and spatial features of biomedical data. Experimental evaluations reveal that the proposed approach produces better accuracy of 98.31% and better precision rate (98.78%). These results demonstrate the potential of the MAYFLY-based CNN-BiGRU model in differentiating PD patients from healthy individuals.
Top1. Introduction
Parkinson's disease (PD) is a neural cells degenerative disease which destroys neurons in the brain's cells. PD mostly impacts people between the ages of 60 which affect both men and women. Dopamine is the neurotransmitter, a chemical messenger in the human brain that gives the “feel-good” sensation because it's associated with pleasure, reward, motivation, and various cognitive processes. A complex combination of elements, including external stimuli, internal states, and neuronal circuitry, is required in the human brain for stimulating dopamine secretion for regulation. Excessive secretion of dopamine leads to addicts like cocaine and amphetamines. However, when the amounts of dopamine in the brain falls, it causes Parkinson's disease symptoms such as tremors, slow movement and stiffness in brain.
The timely identification of Parkinson's disease is vital for various reasons. First and foremost, it facilitates timely intervention, enabling healthcare providers to initiate therapies promptly. Furthermore, early treatment optimization can aid in identifying the most efficacious pharmaceutical regimen, which is particularly crucial considering the inherent diversity in individuals' response to medications. Furthermore, early implementation of therapies such as exercise, physical therapy, and speech therapy can help maintain motor function and avoid potential consequences. Deep learning (DL) and machine learning (ML) methods play a crucial role in the identification and management of Parkinson's disease (PD). These technologies utilize extensive datasets and sophisticated algorithms to reveal intricate patterns and correlations in data that might be challenging for human experts to grasp. Machine learning algorithms are used to develop a mathematical model that analyzes the supplied information and generates predictions and conclusions.
Deep learning is a form of machine learning that use artificial intelligence and neural networks to identify and describe intricate patterns and features in data. These networks utilize a sequence of nonlinear operations on unprocessed input data, enabling them to gather and comprehend intricate connections inside the data. The dataset includes two parts: one is a trained data set which contains of more data, another one is the testing data set. The training data set has 196 observation points and 24 features.
Selected features were extracted from the dataset that will actually affect the PD. This feature extraction is done with may-fly optimisation. Mayfly Optimization is a lower-known nature-inspired optimization method that has been used for a variety of machine learning and optimization problems. It creates an optimization algorithm by drawing inspiration from the mating and reproductive behaviour of mayflies, which are insects with short lifespans. The agents (representing potential solutions) in the Mayfly Optimization algorithm replicate the behaviour of male and female mayflies. The technique iteratively evolves a population of alternative solutions towards an optimal or near-optimal solution using a combination of velocity update, feature selection and crossover/mutation operations.
CNN is employed with RNN (Bi-Directional Gated Recurrent Unit) algorithms for better results with greater accuracy. The word “Convolutional Bi-Directional Gated Recurrent Unit” generally refers to a hybrid model for classification tasks that combines the strengths of CNNs and Bi-GRUs. The model begins with a convolutional layer to process input data and capture local features. The convolutional layer's output is then passed to a Bi-GRU layer, which capture sequential dependencies and context. Finally, for the classification task, the Bi-GRU layer output could be connected to one or more fully connected layers.
The analytical results show greater improvement when compared to previous existing methods with
Train data: Accuracy - 0.9926%, Precision - 0.9833%, ROC_Score (Sensitivity & Specificity) - 0.9953%, F-Score - 0.9892%, MCC - 0.9786%.
Test data: Accuracy - 0.9831%, Precision - 0.9878%, ROC_Score (Sensitivity & Specificity) - 0.9737%, F-Score - 0.9803%, MCC - 0.9614%.
The following content describes the way of the paper is organized. Section 2 discusses related research works, Section 3 describes the PD dataset and introduced architecture with its flow diagram, Section 4 presents the classification and validation procedure of the model, and Section 5 reports on the data analysis they performed on the considered dataset, a discussion of the results and a visualization of our dataset. Finally, Section 6 provides the paper to a conclusion.