Feature selection on imbalanced data set for the decision support of Parkinson’s disease

Abstract

This paper presents results of feature selection on imbalanced data set for clinical decision support of Parkinson’s disease. A Clinical Decision Support system was used to test collected clinical and image analysis data. In total the images of 341 subjects were analyzed: 118 of them had clinical Parkinson's disease diagnosis, 73 healthy controls, 92 had established essential tremor diagnosis, and 24 had mild cognitive impairment. To overcome data imbalance problem for the last diagnosis the Synthetic Minority Over-sampling Technique was used giving the ability to perform feature selection. The number of performed experiments show that used methods help gain a reliable decision. Two features listed in all three diagnosis were selected with importance of 100 % out of 194 gathered during an automated transcranial sonography image analysis. The proposed approach is a supplementary tool for the automated assessment of the parameters for the decision support in the diagnostics of Parkinson's disease.