The definitive treatment of Parkinson's disease, which causes movement disorders and has been increasing in recent years, is still not available today. However, effective studies are being conducted to improve the quality of life of the patients. In this study, a method for the efficient classification of data from Pakinson's disease and normal individuals is proposed. Since the dataset used in the proposed study consists of replicated samples, independence-based classifiers cannot be used for this dataset. When the distribution of the features in the dataset is examined, the success of the classical classifiers is very low due to the fact that the distribution centers between the clusters are very close to each other. Based on the basic idea that increasing the distance of the cluster centers from each other will increase the success, dimensionality techniques such as PCA, ICA, Relieff, RICA have been used. When the desired success was not achieved, a bond theory was established using a two-stage Whale optimization algorithm. Accordingly, the features of the three samples taken from an individual are closed to each other in the feature space, the total samples belonging to the same class are drawn to one side of the feature space and the feature space of the other class is positioned farthest from the center point. Thus, 3 different samples belonging to the same individual will be classified with the same label. In addition, since class difference will be high, classical classifiers such as SVM, k-NN LDA will be able to work successfully. The proposed method is compared with other techniques and as a result, it is seen that the representation ability in the property space is stronger than other related methods.