Most features of health data that have many irrelevant features can reduce the performance of classification method. One health data that has many attributes is the Pima Indian Diabetes dataset and Thyroid. Diabetes is a deadly disease caused by the increasing of blood sugar because of the body's inability to produce enough insulin and its complications can lead to heart attacks and strokes. The purpose of this research is to do a combination of Correlated Naïve Bayes method and Wrapper-based feature selection to classification of health data. The stages of this research consist of several stages, namely; (1) the collection of Pima Indian Diabetes and Thyroid dataset from UCI Machine Learning Repository, (2) pre-processing data such as transformation, Scaling, and Wrapper-based feature selection, (3) classification using the Correlated Naive Bayes and Naive Bayes methods, and (4) performance test based on its accuracy using the 10-fold cross validation method. Based on the results, the combination of Correlated Naive Bayes method and Wrapper-based feature selection get the best accuracy for both datasets used. For Pima Indian Diabetes dataset, the accuracy is 71,4% and the Thyroid dataset accuracy is 79,38%. Thus, the combination of Correlated Naïve Bayes method and Wrapper-based feature selection result in better accuracy without feature selection with an increase of 4,1% for Pima Indian Diabetes dataset and 0,48% for the Thyroid dataset.