APPLYING MACHINE LEARNING METHODS TO ACOUSTIC SIGNAL CLASSIFICATION USING SPECTRUM CHARACTERISTICS

Machine learning methods are modern tendencies in research and processing of any amount of information. Their use is effective for classifying a large number of recorded acoustic signals. Spectral characteristic of a sound signal allows you to carry out research on its frequency components. To assess an effectiveness of machine learning methods was used 3240 acoustic recordings of heart work as an example. Cardiovascular diseases are the most common causes of death of people all over the world. According to the estimates of the World Health Organization, about 17.9 million people die each year from diseases of the cardiovascular system, which is 31 % of all global deaths [1]. There are various methods for detecting pathological conditions of the hear. This methods based on the analysis of sound phenomena that occur during its work. Logistic regression model, a random forest model were used for classification. Also, neural network was trained and tested to solve classification problem. Performance evaluation of models will be carried out according to four metrics (accuracy, precision, recall, F1).

Authors: D. A. Kuzin, L. G. Statsenko, P. N. Anisimov, M. M. Smirnova

Direction: Informatics, Computer Technologies And Control

Keywords: Machine learning, acoustic signal classification, spectral analysis, artificial neural network, logistic regression, random forest

---

View full article