The primary Parkinsons disease (PD) is a disorder of the central nervous system and about 89% of the people with PD suffering from speech and voice disorders. In this project, we adopted a dynamic feature selection based on fuzzy entropy measures for speech pattern classification of Parkinsons diseases. To investigate the effect of feature selection, XGBoost algorithm was applied to distinguish voice samples between PD patients and health people. The data set of this research is composed of voice signals from 195 people, 147 with Parkinsons disease and 48 healthy people. The results show that various voice samples need different feature selection. We applied dynamic feature selection can get higher rate of classification accuracy than all features selected.