Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{199345,
        author = {Akram Hossain and Dr Bimal Dutta},
        title = {Early-Stage Diabetes Risk Prediction using Explainable and Uncertainty-Aware Deep Learning: A Multi-Domain Framework with Selective Prediction and Polynomial Feature Stacking},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {12},
        number = {11},
        pages = {15865-15872},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=199345},
        abstract = {Disease T2DM is an undiagnosed, fast-rising disease prevalent in adults with a worldwide incidence of 537 million individuals; thus, the need to identify a method to predict this disease early and by non-invasive means is a pressing challenge. Nonetheless, current machine learning models have limited generalizability, do not estimate uncertainty, and cannot be applied clinically, which are important research gaps. The research paper is aimed at filling in the gap of uncertain and reliable predicted diabetes systems, and extrapolating on heterogeneous data sets.
This study aims to create a robust and uncertainty-sensitive diabetes prediction model applicable to heterogeneous data and aiding clinical decision-making. The suggested multi-domain framework integrates the data harmonisation realized through MICE with the expansion of the number of features to a polysomnoid (844 features), and uncertainty-aware deep learning and calibrated stacking. The system incorporates uncertainty estimation, calibration, and selective prediction to enhance the reliability of clinical decision-making. Uncertainty on a per-patient basis is offered by a residual neural network with Monte Carlo dropout, and selective prediction is yet another method that helps to increase the trustworthiness of the model.
The results of the experiment show a great improvement, with an AUC of 0.908 (Random Forest = 0.797 and Logistic Regression = 0.735) and high calibration (ECE = 0.031, Brier = 0.128). The given system is intended to be used in real-life screening situations when the reliability of prediction and interpretability are crucial factors. In sum, the current piece of work offers a clinically implementable framework that can be used to promote the screening of diabetes and informed decision-making.},
        keywords = {diabetes prediction, Monte Carlo Dropout, selective prediction, MICE imputation, polynomial features, stacking ensemble, SHAP, uncertainty quantification, domain adversarial learning.},
        month = {April},
        }