부산대학교 도서관

In this study, a novel approach to detect five different stress conditions in a simulated stress database in Hindi language using Interspeech 2010 features are proposed which can be used in psychotherapy in an Indian context to monitor the stress conditions of patients. The features are extracted using an OpenSMILE toolkit. The Naive Bayes, Sequential Minimal Optimization (SMO), Simple Logistic and decision tree classifier are considered for classification. The dimensionality reduction is implemented using an attribute selection technique which makes use of a search method. The accuracy rate achieved using each classifier are tabulated and compared. The random forest decision tree classifier is found to achieve better results. The proposed method achieved a 3 point accuracy improvement without using attribute selection and 1.3 point accuracy improvement with attribute selection when compared with the reported existing work done using the same database.