Prediction of Neonatal Jaundice using Fuzzy Clustering Methods

H.T. Moghaddam, F. Towhidkhah, R. Khayati, M. Torkaman, and Z. Kavehmanesh (Iran)


Fuzzy Clustering, Neonatal Jaundice, Prediction, fuzzy ARTMAP, Fuzzy C-Means(FCM), Subtractive Clustering


Prediction of diseases in most cases is still challenging and unresolved by physicians. Jaundice is the most common and one of the most vexing problems that can occur in the newborns. Although most jaundiced infants are otherwise perfectly healthy, they make us anxious because bilirubin is potentially toxic to the central nervous system and kernicterus can occur. Some different fuzzy clustering methods including fuzzy c-means (FCM), subtractive clustering (SC) and fuzzy adaptive resonance theory mapping (fuzzy ARTMAP) are presented for prediction of the risk of jaundice before and after delivery (in first 48 hours) of newborns. A total of 552 medical records were collected from newborns during April to June 2006 in two general hospitals in Tehran, Iran. To evaluate results of the applied methods we used evaluation performance matrix criteria, which include correct classification (CC%), sensitivity (SE%), and specifity (SP%). The above mentioned criteria for jaundice prediction before delivery were approximately 76%, 97%, 56%, while for jaundice prediction after delivery were 81%, 88%, 67%, respectively. These results show that the proposed systems can achieve satisfying results for predicting risk of jaundice considering this fact that physicians do not have any estimation about probability of jaundice appearance.

Important Links:

Go Back