Early Prediction of Parkinson’s Disease using Multiple SVM Classifiers

Parkinson’s disease is a neurodegenerative condition that primarily affects the motor system of the central nervous system. Slow motions, postural instability, tremors, stiffness, and cognitive impairment are among the motor and non-motor symptoms caused by the progressive loss of dopamine-producing neurons in the substantia nigra area of the brain. Changes in speech and voice, such as a drop in volume or monotonous or incorrect pronunciation, are among the main signs of PD. In its latter stages, Parkinson’s disease can cause impairment and loss of freedom. Early diagnosis and treatment can offer patients and their families more time to plan for the future, including financial planning and making arrangements for long-term care, and can help postpone the development of disability and keep function and independence for longer. Different researchers have used SVM for early PD detection but datasets on PD are frequently unbalanced, with more samples from one class than the other. Unbalanced data may be difficult for a single SVM model to manage, producing biased results. Also, single SVM models can be prone to overfitting, which causes them to perform well on training data but poorly on fresh data. When the training data set is tiny, chances of overfitting increase. In order to resolve this problem, in this study, a collection of biomedical speech measurements is used as a dataset, and an ensemble SVM technique is used to improve the prediction of PD. Feature selection using regression and lasso is performed, followed by data normalization, and parameters of multiple SVM classifiers are trained to work together with selected features and scaled training data, then performance is assessed. SVMs have been utilized as SVM offer feature selection and can locate the features that are most pertinent for classification. Multiple SVMs are resilient to noise and can manage unbalanced data and can also help in reducing overfitting, two issues that are frequently present in medical datasets. The approach utilized in this article produced an enhanced accuracy of 93% in the analysis that followed, as opposed to 87% for the old SVM model using the same data.