Enhancing the Diagnosis of Gestational Diabetes Mellitus: A Comparative Study of Deep Learning and Traditional Machine Learning Models on Imbalanced Datasets

Abstract

This study aims to diagnose gestational diabetes mellitus (GDM) using advanced deep learning and traditional machine learning models, focusing on handling imbalanced datasets. The research objective is to develop models that can accurately classify GDM cases, which are often underrepresented in medical datasets. The methodology involves the use of both deep learning models, such as the Advanced Hybrid Model and Voting Model, and traditional machine learning models, including Random Forest, Gradient Boosting, and LightGBM. These models were trained on a balanced dataset achieved through the Synthetic Minority Over-sampling Technique (SMOTE) and evaluated using cross-validation and test accuracy metrics. The results indicated that the deep learning models achieved high cross-validation accuracy but faced challenges in classifying GDM cases on the test set, with lower precision and recall for the minority class. Traditional machine learning models also demonstrated strong performance but similarly struggled with sensitivity towards GDM cases. The study concludes that while these models show promise in diagnosing GDM, further refinement is necessary to improve their ability to handle imbalanced datasets. Future research should explore advanced techniques to enhance the detection of GDM cases, contributing to more reliable and accurate medical diagnostics.