Hybrid Machine Learning Classifier and Ensemble Techniques to Detect Parkinson’s Disease Patients

Abstract

Parkinson’s disease is caused by tumors, a progressive nervous system disorder that affects development. Stiffness or slow movement is the basic sign of this problem. There is no cure for Parkinson's disease, but some drugs can improve the condition, and sometimes brain surgery can help patients improve their condition. Using machine learning strategies, we developed a priori model to identify patients affected by Parkinson’s disease. By controlling the importance of features, we recognize the most significant indicators of patients who belong to this disease-related estimate. The model-based logic strategies we use include logistic regression (LR), k nearest neighbors (k-NN), support vector classifier (SVC), gradient boosting classifier (GBC), and random forest classifier (RF). The estimated reliability, like the ROC curve and confusion matrix, is five-fold cross-validation. We construct another model that depends on the ensemble method and utilization of majority voting, weighted average, bagging, Ada_boost and Gradient_boosting. The model is also recognized in the five-fold cross-validation and confusion matrix, precision; recall rate and F1 score. The correlation matrix is also drawn to show whether these features are related to each other. Our findings indicate that, compared with different methods, machine learning can provide more reliable clinical outcome assessments for patients with Parkinson’s disease. Among the five algorithms, the higher accuracy fluctuates in the middle of 70–95%. Among them, SVC obtains 93.83% accuracy from the five basic classifiers, and Bagging obtains 73.28% accuracy from the ensemble technique.