Researchers compared five feature selection methods and one dimensionality reduction technique on the UCI thyroid disease dataset, using a combination of Random Forest and XGBoost importance scores to select features. The hybrid feature selection approach with the XGBoost classifier achieved a test accuracy of 0.9842, an average accuracy of 0.9808 across 10-fold cross-validation, and a standard deviation of 0.0063. The highest overall accuracy of 0.9887 was achieved by combining Boruta with XGBoost.
Source:BIOENGINEER.ORG · Read Original
Leave a Reply