This paper aims at analyzing the machine learning techniques for prediction of Permanent Neonatal Diabetes Mellitus (PNDM). PNDM is a serious condition responsible for neonatal mortality. This can be caused due to insufficient insulin levels. The newborn baby in its month after birth is termed as “neonate”. One of the Sustainable development goals provided by world health organization is to reduce neonatal mortality. Machine learning approach for disease prediction is a noninvasive method of dealing with the available data and its interpretation. The main objective of the paper is to assess the best suitable machine learning method for the prediction of PNDM. The dataset used in the analysis consists of six input features as age, information related to genetics, HbA1c level, medical attributes, laboratory details and maternal history. The output feature or target variable in this dataset is the presence or absence of PNDM. [1]. The dataset is divided in training and testing modules in the ratio of 8:2. The dataset is validated and tested using various machine learning techniques such as Decision Tree classifier, Support Vector Machines, Logistic Regression, Naïve Bayes and Ensemble classifier. It was observed that, tree classifier was the best suitable choice for prediction of PNDM as it provided 99% accuracy and lowest training and validation cost as compared to other methods.

Permanent Neonatal Diabetes Mellitus (PNDM) machine learning neonatal healthcare decision tree classifier SVM classifier.