Artificial Intelligence-Based Parkinson’s Disease Identification: A Comparative Analysis of Recurrent Neural Networks and Machine Learning Models

Abstract

The neurodegenerative disorder weakens the brain's neurological, physiological, and behavioural systems; accurate identification is challenging in the early stages due to mild variances. Parkinson's disease is a major health concern. General signs of this condition include sluggish movements called ‘bradykinesia’. Parkinson's disease, a neurological disease, causes tremors, rigidity, imbalance, and loss of motor control. Most symptoms progress over time. This study suggests using supervised algorithms like Logistic regression, SVM, Decision tree, K-NN, Random Forest, Bagging, XGBoost Classifier, ANN /NN and LSTM (part of RNN) to generate the models. to diagnose subjective diseases. The proposed method involves feature selection using filter and wrapper methods, and classification processes. Data for this research were acquired from the UCI Machine Learning Repository. The dataset comprises biological voice recordings from 31 subjects, including 23 patients affected by Parkinson’s disease and 8 healthy individuals. Based on the results of the experiment, it can be stated that XGBoost has the highest accuracy with 97%, followed by KNN with an accuracy 95%, while RNN’s model LSTM achieved 97%, along with 50 epochs. Compared with prior research, the current study's findings indicate that the proposed approach yields equivalent or even superior outcomes.