Newborns' cry analysis classification using signal processing and data mining

Abstract

Newborns' cry is one of the very few indicators on the newborns' state of health. A couple of studies have been performed in the last 20 years with the goal of extracting valuable information from the newborns' cry in order to find out valuable information that would normally be obtained from excessive invasive tests or not available at all with current state of the art medical techniques. Among the focuses of the most recent studies, pathologies like asphyxia, hypoxia, hypothyroidism or hearing disorders are investigated to determine correlations between these and features from the cry signal. In this study, a considerable amount of recorded cry signals from newborns' (300 subjects) is studied in order to classify, by the use of data mining techniques. For this classification four categories of newborns' were considered: newborns' with no detected health issues or problems at birth, newborns' that suffered umbilical cord strangulation at birth, premature born babies (before 38 weeks of pregnancy) and newborns' with different pathologies not included in one of the above categories. By the use of data mining techniques (classification trees, decision rules and lazy algorithms) a classification with very good accuracy has been made for the first three of the above mentioned newborn categories.