A Neonatal jaundice prediction system based on the support vector machine algorithm

Jaundice is one of the most common problems in the neonatal period, affecting about 60% of full-term and 80% of premature infants in their first week of life. The present study aimed to develop a system for predicting neonatal jaundice within the first 24 to 72 hours post-delivery by using the Support Vector Machine (SVM) algorithm.

This applied-developmental study employed a quantitative method. First, based on a literature review, a questionnaire containing effective factors for predicting jaundice in newborns was designed. Data analysis was performed using descriptive statistics, and factors that were recognized as necessary by at least 50% of the experts were included in the model. Then, data from 1178 newborns delivered at Lolagar hospital in Tehran were extracted from birth records, and several machine learning algorithms were used to predict neonatal jaundice.

The findings of this research showed that the proposed model based on the SVM algorithm is the best output due to the distance between classes. Therefore, the final model of the SVM algorithm was created using the Gaussian kernel, with a sigma value of 1.2360605. Thirty percent of the samples (354 cases) were tested, and 321 cases were correctly predicted. In the proposed SVM model, parameters such as precision, the area under the Receiver Operating Characteristic (ROC), and F1 score were 92.7%, 93%, and 88% respectively.

Incorporating SVM into a system for predicting jaundice in newborns can aid doctors with timely prediction of jaundice in newborns and provide the possibility of taking preventive measures and preventing possible risks caused by jaundice in newborns.