Parkinson’s disease classification using nature inspired feature selection and recursive feature elimination

New progress in machine learning (ML) have paved the way for solving various existing complex problems, including in the medical field. In the proposed paper, we have investigated methods to detect Parkinson’s Disease (PD) with Nature Inspired Feature Selection (NIFS) using the Zebra Optimization Algorithm and Recursive Feature Elimination Cross Validation (RFECV). A vocal feature-based dataset has been used for PD Detection as it contains data relating to the vocal features of PD patients. It has been proven from research that vocal disorders are observed in the majority of individuals in the preliminary phases of this disease. The number of features has been reduced from 754, in the original dataset, to 40 using feature selection, and the classification results have been obtained for two cases, one being a 70:30 train test split and the other being tenfold cross-validation. We have implemented the proposed technique on 11 different classifiers. Out of these, the Gaussian Process classifier showed the best accuracy for both cases. The accuracy values obtained for cases one and two are 96% and 97.07%, respectively, one of the highest accuracy values obtained for similar research done by other researchers. Additionally, the generalization capability of the model obtained may be enhanced by including more data points in the dataset.