An Analysis of Deep Learning and Machine Learning Approaches for Diabetes Mellitus Prediction in Pregnant Women

In our country, one significant public health concern is diabetes mellitus. Thousands of people suffer from this metabolic disease, which is caused by the body accumulating too much glucose. Diabetes can lead to several health problems in humans if it is not identified early on, including kidney issues, nerve damage, vision damage, heart disease, and more. Gestational diabetes mellitus (GDM) is the name for diabetes mellitus that occurs during pregnancy, both the mother and the unborn child or fetus are at serious risk from it. For prompt intervention and better results for the mother and fetus, early and precise GDM prediction is essential. This study aims to predict diabetes mellitus in pregnant women using a real-time diabetes dataset, machine learning algorithms including Random Forest, Gradient Boost, Decision Tree, and XG-Boost classifiers alongside deep learning algorithms like Long Short-Term Memory (LSTM), Recurrent neural network (RNN), Gated Recurrent Unit (GRU), and hybrid GRU+LSTM. In this work, a detailed analysis and methods such as the Synthetic Minority Oversampling Technique (SMOTE) are used to produce the best probable answer for the prediction of diabetes mellitus. The Pima Indian Diabetes Dataset from the National Institutes of Diabetes and Digestive and Kidney Diseases is used in this investigation. All 768 patients are female and at least 21 years old. This dataset contains one target variable and eight predictor variables. It has been discovered that XG-Boost (82.47%) offers the highest accuracy after comparing the accuracies of other algorithms like Gradient Boost (81.82%), Random Forest (81.17%), Decision Tree (66.23%), GRU (79.69%), LSTM (77.08%), RNN (81.25%) and hybrid GRU+LSTM (80.21%) classifiers using imbalanced datasets (without using SMOTE algorithm). This study examines the model accuracies, after applying SMOTE algorithm, Random Forest algorithm, stood top with accuracy such as 88% surpassing the accuracies of remaining models XG-Boost (85.2%), Gradient Boost (84.4%), Decision Tree (83.6%), GRU (80.8%), LSTM (78%), RNN (80.4%) and hybrid GRU+LSTM (80.8%). This paper compares the existing model like ensemble model accuracy of 86% with the proposed model accuracies, and the proposed model has been observed to outperform the accuracy with 88% in Random Forest. The proposed work uses the Application Programming Interfaces using Flas. Flask is like a simple toolbox for making websites with Python.