Modeling Voice Pathology Detection Using Imbalanced Learning

To solve the problem of imbalanced data distribution of MEEI database, a modeling voice pathology detection method use imbalanced learning algorithm is presented in this paper. The method is based on support vector machine (SVM), decision tree (DT) and random forest (RF) to build voice pathology detection(VPD) model. Three imbalanced learning algorithms, SMOTE, Borderline-SMOTE and ADASYN, oversample a minority class (normal voice samples) and finally a training set of class-balanced is obtained. 10-fold cross validation is used in the experiment, and Precision, Recall and AUC/PRC are selected as model evaluation measures. Experimental results show that the imbalanced learning algorithm improves the recognition ability of pathological voice detection model for a minority class. RF as an ensemble model can combine better with imbalanced algorithms and performance slightly better than single classifier models. Its originality lies in the consideration of MEEI database as training set of pathological voice model, the effect of class-imbalanced on model performance, and Accuracy (Acc) as the main evaluation measure of VPD model is not suitable for training set of class-imbalanced.