Predicting Parkinson's: Advancing Early Detection Through Machine Learning

In today's world, health has become a paramount concern. Parkinson's disease, a progressive neurodegenerative disorder affecting the nervous system and areas of the body controlled by nerves, afflicts over one million people worldwide, ranking as the second most prevalent neural disease. Leveraging predictive models, healthcare professionals can make well-informed decisions, enhance patient care, and ultimately improve the quality of life for those with Parkinson's disease, potentially saving 1–2% of lives. The primary focus of this study is to develop a model for early Parkinson's disease prediction. The dataset used for analysis contains clinical assessments along with voice recordings of patients and exhibits high dimensionality. Additionally, it's an imbalanced dataset. To address the dataset's imbalance, employed the Synthetic Minority Over-sampling Technique (SMOTE). For feature selection, correlation values between the features and the target variable. Specifically, reduced the feature set from 754 to 350 by eliminating features with negative correlation values with the target variable and as part of data preprocessing, applied stratified k-fold cross-validation and min-max scaling to enhance accuracy. This study compares four widely employed classifiers: Support Vector Machine (SVM), RandomForest, XGBoost, and CatBoost. The experimental results obtained show that the proposed model with the CatBoost classifier has performed better when compared to the remaining three classifiers. The proposed model has obtained an accuracy of 93.39% after performing classification using the CatBoost classifier. Moreover, the performance of the proposed model is compared with the five state-of-the-art techniques in which the proposed model outperforms all the techniques.