Akademik Veri Yönetim Sistemi
Gazi University Journal of Science, cilt.39, sa.1, ss.113-128, 2026 (ESCI, Scopus, TRDizin)
Voice is produced by excited dynamic vocal tract system consisting of vocal cords, lips, tongue, with the air from lungs. Pathological cases in these components change the characteristics of the voice. Voice diseases impair quality of life, and some pathologies can be life-threatening, so early diagnosis is important. In literature, voice disorders can be generally investigated with traditional or acoustics signal analysis methods and detected by using machine learning. In this study, analysis and classification studies were realized for detecting vocal cord deformities such as cyst, polyp and sulcus vocalis, and healthy ones. Unlike literature, instead of acoustic features, features based on information theory such as recurrence plot quantities, entropies and dimensions coming from nonlinear approaches were used. Three feature selection and three classification methods were tried with nonlinear features calculated from whole signal and decomposed signal with discrete wavelet transform. Two classification procedures (binary and multiple class) with 10-fold cross validation were applied and tested with distinct groups. According to the results, in binary classification (healthy/ diseased), the best test accuracy of 99.2% with Coarse Tree classifier was obtained with only four features selected with MRMR algorithm. In multiclass (healthy/cyst/polyp/sulcus), the test accuracy of 91.6% was found as the best with only eight selected features by using ANOVA and Kernel Naive Bayes classifier. The results show that the use of a few features coming from nonlinear domain provided very effective and successful in classifying voice disorders in both binary and multiclass when compared to the studies in the literature.