Comparison of Feature Selection in The Classification Of Patients With Parkinson's Disease

Parkinson's disease is a weakened disease brain function so that early treatment so far has not been able to be done surgically. Parkinson's disease is a disease that involves a degenerative process in dopaminergic neurons in the substantia nigra. Some parameters for the detection of Parkinson's disease are not continuous, so the role of expert doctors cannot be ignored. Therefore, to help doctors, research in the field of machine learning is done on the detection of patients with Parkinson's disease. This research discusses comparisons of the classification techniques of patients with Parkinson's disease using three feature selection methods, namely information Gain (filter method), Forward Selection (wrapper method) dan Maximum Relevance Minimum Redundancy (mRMR). Feature selection with mRMR approach can improve accuracy compared to filter method and Wrapper method (
$IG < FS < mRMR$
). Filter method is no better than a wrapper method in experiments conducted. The results show that attribute weighting with mRMR is better than the other two methods.