Disease Detection in Voice Data by Using Machine Learning Algorithms

Abstract

Late detection of diseases reduces recovery rates, complicates the treatment process, and increases the cost of recovery. Therefore, early detection of diseases is important. Nowadays, machine learning, artificial intelligence, deep learning methods are widely used in the field of health for medical data analysis and disease detection purposes. Hoarseness is a common complaint in the community. Various diseases such as vocal cord polyp, laryngeal cancer, acute laryngitis, vocal cord paralysis can cause hoarseness. While hoarseness is a common symptom among these diseases, their causes, treatment processes, and risks vary. In this thesis, the aim was to make a basic classification of different diseases causing hoarseness using voice data before pathological and endoscopic examination. The study utilized the audio data obtained from the Saarbruecken database created by the Institute of Phonetics at Saarland University. Classification was performed using five different machine learning algorithms: K Nearest Neighbor, Naive Bayes, Decision Tree, Support Vector Machine, Random Forest on a total of 652 voice data from patients diagnosed with Reinke's edema, laryngitis, cancer, polyps, vocal cord paralysis, and healthy individuals. The results obtained were compared using a confusion matrix.