Ensemble of Boosting Algorithms for Parkinson Disease Diagnosis

Parkinson’s disease (PD) is a common dynamic neurodegenerative disorder due to the lack of the brain’s chemical dopamine, impairing motor and nonmotor symptoms. The PD patients undergo vocal cord dysfunctions, producing speech impairment, an early and essential PD indicator. The researchers are contributing to building generic data-driven decision-making systems due to the non-availability of the medical test(s) for the early PD diagnosis. This article has provided an automatic decision-making framework for PD detection by proposing a weighted ensemble of machine learning (ML) boosting classifiers: random forest (RF), AdaBoost (AdB), and XGBoost (XGB). The introduced framework has incorporated outlier rejection (OR) and attribute selection (AS) as the recommended preprocessing. The experimental results reveal that the one-class support vector machine-based OR followed by information gain-based AS performs the best preprocessing in the aimed task. Additionally, one of the proposed ensemble models has outputted an average area under the ROC curve (AUC) of 0.972, outperforming the individual RF, AdB, and XGB classifiers with the margins of $$0.5\,\%$$ , $$3.7\,\%$$ , and $$1.4\,\%$$ , respectively, while the advised preprocessing is incorporated. Since the suggested system provides better PD diagnosis results, it can be a practical decision-making tool for clinicians in PD diagnosis.