Handling Uncertainty in Parkinson’s Disease Voice Data Using Intuitionistic Fuzzy Entropy Measure

One of the challenges during development of machine-learning model is to deal with the uncertainty of data. The performances of machine-learning algorithms are affected by of uncertainty sources that are epistemic and aleatoric. IFS (Intuitionistic Fuzzy Set) entropy deal with the uncertainty of features present in Parkinson’s disease data. Feature selection actually performs the reduction in dimensionality of features which augment the performance of classification. A method of feature selection depending on IF Entropy measure is investigated here for finding features of voice data for development of machine-learning model for Parkinson’s disease detection. Features with less uncertainty or entropy (high information) values are chosen to classify the Parkinson’s disease (PD) patients from Healthy Control (HC). Experiments are conducted on PD patient’s voice datasets to find the efficiency of the proposed scheme and three Machine-Learning (ML) techniques, i.e., K Nearest Neighbor (KNN), Support Vector Machine (SVM) and Ensemble Bagged Tree (EBT). The proposed Intuitionist Fuzzy set based entropy based selection shows impressive results as comparatively better results are obtained only with as low as 27% features (only 6 in 22 numbers of features) for KNN and EBT while 36% features for SVM, while reducing features with in order of higher uncertainty values.