Authors: Tae-Il Kim, Hee-Jung Park, Eung-Hee Kim, Harshit Kumar and Hong-Gee Kim
Abstract: Determining the factors of disease and analyzing its pattern is one of the key issues for current healthcare research. Xerostomia is caused by lack of saliva secretion and often results in oral inflammation including periodontal disease. The purpose of this study was to build a model that can predict deciding factors related to xerostomia. We performed with two methods (Logistic Regression, LR; Support Vector Machine, SVM) using the questionnaire. Two groups were constructed; one was comprised of healthy subjects and the other was xerostomia suffering group. After the data was balanced, sequential forward search (SFS) algorithm with SVM was applied. In order to compare the predictive abilities of both methods, sensitivity, specificity, and positive/negative prediction value were obtained from 10-fold cross validation procedure. The results indicated that the classification accuracy of SVM with SFS was higher than that of LR, and the discriminative ability of SVM revealed outstanding performance comparing to LR for every criterion. In conclusion, SVM would be an effective method for xerostomia prediction procedure.
Keywords: Logistic models; Oral health; Support vector machines; Xerostomia