Parkinson's disease (PD) arises from the degeneration of neurons and the subsequent depletion of dopamine, resulting in symptoms such as tremors, muscle rigidity, and bradykinesia. Timely identification is crucial; however, existing techniques do not offer a conclusive remedy. This work aims to fill the existing gap by utilizing open-source Python-trained models to evaluate the potential of auditory data in classifying Parkinson's disease, applying a range of machine learning algorithms, such as neural networks, logistic regression, random forest, adaboost, and k-nearest neighbors, to the UCI telemonitoring dataset, which consists of 31 persons, including 23 with Parkinson's disease. The evaluation is done using parameters including accuracy, precision, and recall. The suggested framework prioritizes data preprocessing, segmentation, algorithm training, and comprehensive evaluation, highlighting the significance of data preparation and algorithmic assessment in predictive modelling for early identification of Parkinson's disease.
Top1. Introduction
The movement of the body's muscles is significantly impacted by Parkinson's disease (PD), a common neurological illness (Max A. Little, 2008). It results in tremors, muscle rigidity, and bradykinesia, as well as changes in posture, speech, and mobility (Prabhavathi, K, 2022). Rooted in the demise of neurons, PD diminishes dopamine levels within the brain. This deficiency hampers synaptic communication, leading to compromised motor function (Braak, H., 2000,). The spectrum of symptoms arising from the demise of dopaminergic neurons encompasses a range of manifestations, with tremors and balance irregularities being the most prevalent. Notably, symptom progression can vary markedly among patients. To stave off the relentless progression of PD, patients must depend on early detection and personalized therapeutic interventions, as a definitive cure remains elusive. There has been relatively little research on hearing impairments as a way of early identification, despite the fact that previous literature has made progress in predicting PD using methods including MRI scans, gait analysis, and genetic data. For instance, (Alatas Bilal,, n.d.,), (P. Raundale, C, 2021,) and (Nusinovici, S., 2020,) used a support vector machine (SVM) model to analyse genetic data, and they were able to predict the beginning of PD in older people with an accuracy of 0.889. The accuracy of the revised SVM model in this study is 0.9183, illustrating the advantages of employing auditory data for PD classification over genetic data. Using the UCI telemonitoring dataset, (P. Raundale, C, 2021,) trained a Random Forest classifier using keyboard data to evaluate the severity of PD in elderly adults. While largely reliant on MATLAB, (Sharma, N., 2023), (Maliha, S. K., 2019,) and (Srivastav, G., 2022,) on the other hand, concentrated on auditory data for categorising persons with PD. In this study, we use open-source Python-trained models that offer increased speed and efficiency. Central to this discourse is the application of machine learning techniques in the nascent stages of PD. The dataset encompasses a comprehensive array of biological voice measurements gleaned from 31 participants, including 23 afflicted by PD. The “status” column in the dataset designates a value of 0 for healthy patients and 1 for people with Parkinson's disease (PD), with the latter being a key objective of the dataset. The range of machine learning techniques taken into consideration includes neural networks, logistic regression, random forest, adaboost (Desai, M., 2021; Murugan, A., 2019; Bakare, Y. B., 2021), and K-nearest neighbours (KNN). The efficacy of these algorithms is gauged through diverse metrics, spanning accuracy, precision, and recall, offering a comprehensive evaluation of their performance (Dietterich, T. G., 2000; Pahuja, G., 2021).