Parkinson's Disease Data Analysis and Prediction Using Ensemble Machine Learning Techniques

Parkinson's disease is an incurable neurodegenerative disorder which causes problems with communication, thinking, behavioral problems, and other difficulties. Detecting the disease at an early stage can help uplift the living quality of the patient. This paper presents the findings of data analysis performed on patients' speech signals measurement to understand the insights and uses ensemble models (GBDT, random forest, voting classifier, and stacking classifier) to predict whether a patient is suffering from Parkinson or not. Voting and stacking models use KNN, logistic regression, SVM, random forest, and GBDT as base learners. It can be noted that fundamental frequencies (minimum and average) have higher values for positive cases. Four metrics, F-1 score, accuracy, log loss, and ROC score, were used to report the performance of the model. The best model was SVM which out performed ensemble models in three out of the four performance metrics (F-1 score, accuracy, log loss). However, the stacking and voting ensemble models had better ROC scores than that of the base learners.