Noviyanti Santoso, Santi Wulan Purnami : Comparison Parametric and Non-parametric Classification Methods on Medical Data
Noviyanti Santoso S.Si., M.Si.
Santi Wulan Purnami S.Si., M.Si
The 6th Annual Basic Science International Conference
Discriminant analysis and logistic regression is a parametric statistical method used to solve classification problems. Parametric methods are still tight on the fulfillment of assumptions such as normality and homogeneity. Therefore, the application sometimes has trouble because the data does not meet these assumptions. On development, the current classification method is more modern, a neural network is a machine learning-based classification method that does not require that the data meet the assumptions of any kind, while other methods are also being developed is MARS. This method both is using a non-parametric approach to classify the object. The purpose of this study was to compare the accuracy of the classification using parametric approach discriminant analysis and logistic regression with a non-parametric approach to neural network and MARS. A good method is a method that produces a high classification accuracy or misclassification small. Using data from the UCI Machine Learning is Wiscosin Breast Cancer and share them in the training set and testing set (80:20) the results obtained is a classification accuracy of discriminant analysis of data for training and testing respectively 91.7% and 71.24% while the training data for the logistic regression 92.2% and 71.74% for the data testing. While using a non-parametric method of neural network and MARS, classification accuracy was 91.8% and 75.82% for the testing data. It show that non-parametric methods are more accurate than parametric classification methods on this case.