Type-2 Diabetes Detection using Snow Ablated Graph Networks

• Published in: 2025 5th International Conference on Soft Computing for Security Applications (ICSCSA) pp. 2012 - 2018
• Main Authors: R, Evangeline Kiruba, S, Sophia
• Format: Conference Proceeding
• Language: English
• Published: IEEE 04.08.2025
• Subjects: Accuracy; Brain modeling; Cleaning; Convolution; Diabetes; Disentangled Cascaded Graph Convolution Network; Median & Median Absolute Deviation; Optimization; Orchard Algorithm; Real-time systems; Redundancy; Reliability; Snow; Snow Ablation Optimization and Type-2 Diabetes
• DOI: 10.1109/ICSCSA66339.2025.11171284

The rising prevalence of diabetes, especially in emerging nations, emphasizes the urgent need for accurate and early diagnosis. However, challenges such as limited availability of biomedical data, inconsistencies in health records and manual prediction difficulties hinder effective deployment of reliable diagnostic systems. To overcome this challenge, this research proposes Disentangled Cascaded Graph Convolution Network with Snow Ablation Optimization (D-CGCN-SAO) approaches to improve detection accuracy of Type-2 Diabetes. Initially, Electronic Health Records (EHRs) are collected from PIMA Indians Diabetes Dataset (PIDD), which contains structured medical data of female patients. Subsequently, applying Median & Median Absolute Deviation (MMAD) Normalization to the data for standardization and cleaning. After cleaning, important features are selected using Orchard Algorithm (OA), which reduces redundancy and keeps only most relevant indicators. Then, Disentangled Cascaded Graph Convolution Network (D-CGCN) for enhanced detection. Finally, Snow Ablation Optimization (SAO) technique simulates melting and sublimation dynamics of snow in order to optimize the model's weights. A model is proposed with an accuracy of 99.77%, precision of 99.49% error rate of 0.4% is able to detect the brain tumor with a high reliability for medical diagnosis while minimizing the number of false positives. Compared to existing methods, achieves 99.92% accuracy, 99.54% sensitivity, and 99.45% F-Measure, surpassing current methods by fusing nature-inspired optimization with graph-based feature learning to improve computational efficiency, accuracy, and feature relevance.This shows better performance compared to current methods. The model improves detection accuracy and computational speed.