Integration of ResNet-50 with Adaptive Simulated Annealing for Enhanced Predictive Modeling of Gestational Diabetes

Abstract

Chronic diseases such as cancer, diabetes, and cardiovascular disorders pose significant public health challenges due to their chronic duration and the costly nature of treatment. Gestational diabetes is a complex, costly, and long-term public health issue, particularly during pregnancy, necessitating early diagnosis and precise prognosis to improve both the mother's and the fetus's health. Early diagnosis and effective prediction of disease progression are essential for lowering healthcare costs and enhancing patient outcomes, but they often lack precision and adaptability. The research aims to develop an advanced Artificial Intelligence (AI)-based framework for the early diagnosis and predictive modelling of pregnancy complication datasets using gestational diabetes that includes clinical parameters, physiological biomarkers, and basic demographic information. The dataset is pre-processed by handling missing values and standardizing numerical features to ensure data quality and consistency. Recursive Feature Elimination (RFE) is applied for informative feature extraction. A modified ResNet-50 Deep Learning (DL) architecture is employed for both classification and early diagnosis tasks, while Adaptive Simulated Annealing (ASA) optimizes feature selection. The ResNet-50-ASA model achieved high accuracy (0.985), precision (0.969), sensitivity (0.889), specificity (0.865), F1-score (0.975) and Recall (0.983), enabling accurate early detection and prediction of gestational diabetes. The ResNet-50-ASA architecture outperformed conventional models, achieving high classification accuracy and robust generalization across validation datasets. The model demonstrated improved sensitivity in identifying early signs of chronic diseases like gestational diabetes. The proposed AI-driven pipeline effectively enhances early diagnostic capabilities and predictive modelling for chronic gestational diabetes. Incorporating advanced feature selection and DL techniques offers a promising direction for clinical decision support systems and proactive healthcare interventions.