Diabetes Prediction using Deep Learning: An Analysis of the Pima Indian Dataset

¹ Dept. of Electronics and Communication Engineering, St.Joseph’s College of Engineering, Chennai, Tamil Nadu, India
² Dept. of Electronics and Communication Engineering, St.Joseph’s College of Engineering, Chennai, Tamil Nadu, India
³ Senior System Quality Analyst, TELUS, Portsmouth, United Kingdom

Abstract

Diabetes mellitus is a significant global health concern, and its early and correct diagnosis is required to prevent significant health-related complications in the long run. It is composed of the Pima Indians Diabetes Dataset, which is used to deepen the understanding of diabetes prediction with the help of the deep learning algorithm. There is a lot of preprocessing, including the feature selection through the Extra Trees Classifier, the normalization, the treatment of the missing records or the zero values, and the elimination of the duplicate records. It uses three deep learning models as base learners, namely Convolutional Neural Network (CNN), Long Short-term Memory (LSTM), and Artificial Neural Network (ANN). Key metrics are used to estimate their performance in the aspect of predictive strength. In this strategy, an Ensemble Deep Learning method aims at increasing the credibility of such models by layering their results on top of a meta-level classifier. Experimental evidence demonstrates that the stacked ensemble is better than individual models in every instance and exploits the complementary advantages of the latter. The work demonstrates how ensemble deep-learning may facilitate the creation of powerful clinical decision-support systems to make early diagnosis and prevention of diabetes.