A Mortality Prediction System for Neonatal Jaundice Using Machine LearningTechniques (Preprint)

BACKGROUND Birth complications, especially jaundice, are a leading cause of child death and morbidity across the world. The severity of these diseases may decrease if researchers learn more about their origins and develop effective treatments. Certain advancements have been made, but they are insufficient. Newborns often have jaundice as their primary medical issue. Jaundice may be brought on by a variety of factors. An elevated bilirubin level is a hallmark of jaundice. The incidence of hyperbilirubinemia in newborns is highest during the first postnatal week. The inability to detect problems early enough to get prompt treatment, as well as the similarity of symptoms that may lead to misdiagnosis, are both potential causes of failure. The situation is far worse for Ethiopia and other countries already in distress. A lack of paediatricians and neonatologists might be a reason for alarm. Due to a lack of appropriate diagnostic tools, experts in newborn health are often forced to rely their judgements only on interviews. It's probable the interviewer didn't know much about contagious diseases in infants. This suggests there is room for a faulty or insufficient diagnosis. For machine learning to make accurate forecasts, sufficient amounts of relevant past data must be made available. Jaundice has a high mortality rate, however this may be reduced with prompt identification and classification. The diagnostic accuracy of illnesses may be enhanced by using machine learning techniques. In this essay, I do a deep dive into medical data mining and pull out all the stops to provide you the information you need. It is necessary to investigate, analyse, extract, choose, and categorise the characteristics. Finally, it offers some therapeutic ideas. It helps the doctor diagnose jaundice faster so that effective therapy may be started sooner. The procedure is simplified and made more natural with the use of computer vision and machine learning methods. The refined method of classification improves accuracy. Using a classification stacking method, we found that the top causes of mortality in newborns include serious infections, birth asphyxia, necrotizing enterocolitis, and respiratory distress syndrome. Most infant fatalities may be traced back to these three factors. Dates included in the data set are 2018 through 2021. Support Vector Machine (SVM) performed best when pitted against the newly developed stacking model, XGBoost (XGB), Random Forest (RF), and other machine learning models. The proposed stacking model performed better than its competitors in terms of accuracy (97.04 percent). This is important because we hope it will help hospitals, particularly those with less resource, detect infant diseases sooner. OBJECTIVE A Mortality Prediction System for Neonatal Jaundice Using Machine LearningTechniques METHODS Model Development After selecting an appropriate algorithm, the resulting model may be trained. When developing an AI/ML model, it is important to consider the following: a. Tuning of Hyperparameters By adjusting hyperparameters, data scientists may tailor the performance of machine learning algorithms to a given dataset. A hyperparameter is a parameter whose value controls the learning process. Training is controlled by a set of parameters called hyperparameters. For instance, while configuring a neural network, one must determine the number of hidden layers of nodes to use between the input and output layers, as well as the number of nodes that can fit in each layer. There is zero correlation between these factors and the data used for training. Variables used in setting up a programme. Keep in mind that hyperparameters tend to stay the same even while parameters change throughout training. The hyperparameters are adjusted by the practitioner during the model's configuration. b. Error Analysis By doing an error analysis, we may better understand why your system made inaccurate classifications in some circumstances. This might help decide which problems need greater focus, and how much. It teaches data scientists the proper way to deal with errors. [16] RESULTS Using data from a tertiary hospital in south-western Nigeria, this research created a classification model for predicting the severity of baby jaundice given the values of variables. The dataset included 23 individuals and their predefined cohorts of infants with jaundice, as well as 23 covariates. A deep learning employing MLP classifier trained on an annotated data set was used to categorise the severity of baby jaundice. The accuracy of prediction models built by a classifier was tested after being trained using 10-fold cross-validation. Since the multi-layer perceptron method successfully distinguished between Liver and non-Liver instances, it triumphed over the support vector machine. A deep learning multilayer perceptron method using an MLP classifier with 5 epochs was built after a prediction model for the severity of newborn jaundice was developed. By including the likelihood of Kernicterus and liver disease into the neonatal jaundice prediction model, clinicians may better assess their patients' conditions in real time and make more informed treatment choices. CONCLUSIONS Using data from a tertiary hospital in south-western Nigeria, this research created a classification model for predicting the severity of baby jaundice given the values of variables. The dataset included 23 individuals and their predefined cohorts of infants with jaundice, as well as 23 covariates. A deep learning employing MLP classifier trained on an annotated data set was used to categorise the severity of baby jaundice. The accuracy of prediction models built by a classifier was tested after being trained using 10-fold cross-validation. Since the multi-layer perceptron method successfully distinguished between Liver and non-Liver instances, it triumphed over the support vector machine. A deep learning multilayer perceptron method using an MLP classifier with 5 epochs was built after a prediction model for the severity of newborn jaundice was developed. By including the likelihood of Kernicterus and liver disease into the neonatal jaundice prediction model, clinicians may better assess their patients' conditions in real time and make more informed treatment choices.