Among the main living conditions in newborns is jaundice, characterized by yellowing of the skin due to an abnormal increase in bilirubin in the blood. It is estimated that approximately 60% of full-term newborns and 80% of premature babies have some degree of jaundice in their first weeks of life. Therefore, the objective of this research was to develop a model for the detection of liver disease based on jaundice in newborns that allows distinguishing between patients with and without signs of jaundice using convolutional neural networks and machine learning techniques. The methodology consisted of five stages: acquisition of a dataset; preprocessing (normalization and extraction of the green channel and generation of statistical features); machine learning (RF and XGBoost, SVM, KNN); implementation of CNN architectures (Efficient Net and ResNet50). The results show that the ResNet50-based model achieved an accuracy of 92.1% and a ROC curve of 0.77, confirming its discriminatory ability. These metrics demonstrate that deep neural networks with transfer learning are more effective in identifying jaundice. In conclusion, the results demonstrate that it is possible to apply artificial intelligence in clinical contexts, promoting the development of accessible, scalable, and low-cost medical tools with potential application in areas with limited access to specialized health services.