keyboard_arrow_up
Machine Learning Model to Predict Birth Weight of New Born using Tensorflow

Authors

S.Karthiga, K.Indira and C.V.Nisha Angeline, Thiagrajar College of Engineering, India

Abstract

Low Birth Weight is the major problem for the new born. Low birth weight is a term used to describe babies who are born weighing less than 5 pounds, 8 ounces (2,500 grams). Low-birth weight babies are more likely than babies with normal weight to have health problems as a newborn. Almost 40 percent of the new born suffer from underweight. Predicting birth weight before the birth of the baby is the best way to help the baby get special care as early as possible. It helps us to arrange for doctors and special facilities before the baby is born. There are several factors that affect the birth weight. Through past studies, it has been observed that the factors which affect the child birth range from biological characteristics like the baby's sex, race, age of mother and father, weight gained by the mother during pregnancy to behavioral characteristics like smoking and drinking habits of the mother, the education and living conditions of the parents. This project focuses on developing a web application that predicts baby weight taking baby’s gender, plurality, gestation weeks and mothers age as inputs. Machine learning is one of the domains that plays important role in medical industry. Many machine learning models have been developed to predict diseases at the early stage. In this project wide and deep neural network model is developed using TensorFlow library in Google cloud environment. Wide and Deep Neural Network combines wide linear model and deep neural network. It provides both memorization and generalization. Pre-processing and training is done in the distributed environment using cloud Dataflow and Cloud ML Engine. The model is then deployed as REST API.A web application is developed to invoke the API with the user inputs and show the predicted baby weight to the users. It is scalable and provides high performance.

Keywords

Full Text  Volume 9, Number 15