An Optimized Deep Fusion Network for Early Prediction of Type 2 Diabetes Using Clinical Data
Contributors
Dr. Gunjan Mittal Roy
Keywords
Proceeding
Track
General Track
License
Copyright (c) 2026 Sustainable Global Societies Initiative

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
Abstract
Early identification of individuals at risk of Type 2 Diabetes (T2D) can support timely intervention and reduce the likelihood of disease progression and associated complications. This paper presents a compact hybrid deep-learning framework, termed Gated Bidirectional Grey-Wolf Optimized Deep Fusion Network (GBR-GWO-DF-Net), for T2D classification using structured clinical data. The framework combines gated recurrent feature learning, bidirectional temporal representation, deep feature fusion, and Grey Wolf Optimization (GWO) for hyperparameter optimization. Principal Component Analysis (PCA) is employed as a preprocessing step to obtain a compact and less correlated representation of the clinical variables. The proposed approach is evaluated using the PIMA Indians Diabetes dataset containing eight clinical and demographic predictors and a binary diabetes outcome. Experimental results show an accuracy of 99.42%, sensitivity of 99.31%, specificity of 99.48%, F1-score of 99.17%, precision of 99.03%, and AUC of 0.9998. Ten-fold stratified validation further produces very small standard deviations across the evaluated metrics, indicating stable performance across the validation folds. SHAP-based analysis identifies glucose, BMI, Diabetes Pedigree Function, age, and number of pregnancies as the most influential variables. Comparative experiments against conventional and ensemble approaches demonstrate the potential of the proposed framework for data-driven early T2D risk assessment. Nevertheless, external validation using larger and more heterogeneous clinical populations is required before clinical deployment.