THE EFFECT OF NAIVE BAYES CLASSIFIER USING DUMMY VARIABLE AND FEATURE BACKWARD SELECTION WITH PEARSON CORRELATION IN DIAGNOSING GYNECOLOGY

Abstract

The use of pearson correlation and backward selection as feature selection can be combined to improve the accuracy of the Naive Bayes Classifier. Feature selection is done as preprocessing of data on the process Classfier Naive Bayes algorithm. Pearson Correlation as a preprocessing data will work to sort the parameters that influence the classification process of the gynecological disease and Backward Selection will select those parameters in sequence on the preprocessing data in the Naive Bayes Classfier process. Previously, the parameter would be converted to dummy variables because the parameter has the possibility of a value that appears more than two (non-binary). This study discusses the effect of Naive Bayes Classfier using dummy variables and feature backward selection with pearson correlation in the diagnosis of gynecological disease. The results obtained in this study prove that the use of dummy variables increases the accuracy value from 88% to 88.8% and the use of pearson correlation as preprocessing data increases the accuracy value of Naive Bayes Classifier from 88.8% with 24 parameters to 89.6% with 20 parameters. The use of pearson correlation not only improves accuracy but also increases the effectiveness of the features used in the Naive Bayes Classifier process. This can be seen from the increase of accuracy results with the use of the number of parameters that decreased.