Consequently, we propose a hybrid system combining ANN and SVM classifiers (SVM is tested with two different kernels). The goal is not only to establish a comparison between all of them but also to benefit from the highest accuracies of each classifier. The diagnosis must be reinforced and complemented in order to provide a better generalization in the same way that two or more specialists (or a specialist group) co-operate with each their methods, in order to obtain a final common diagnosis. As illustrated in Figure 5, our system allows finding out which instances are correctly or incorrectly classified. A future line of the system is an exhaustive study of all the fields, thus allowing us to determine why the errors occurred, and learning how to avoid this from happening in the future.
We have found that the outliers and the imbalanced data directly affected the classification performance and effectiveness of the classifiers. There are 147 registers with PD and 48 healthy ones. The accuracy of the classifiers will be improved by eliminating a number of outliers from both the minority and majority classes, and increasing the size of the minority class to the same size of the majority class.