The express growth of Internet of Things (IoT) and Artificial Intelligence (AI) technology is transforming healthcare by allow academic system for early and correct sickness diagnosis. However, existing indicative approaches often effort with difficult medicinal data and hyperparameters tuning, limiting their analytical accuracy and generality. To address this challenge, the researches propose a novel Scalable Whale Optimizer-tuned Graph Attention Network (SWO-GAN) model planned to advance disease diagnosis competence and reliability. Medical datasets such as hepatitis, liver tumor, and Parkinson’s disease was sourced from Kaggle to evaluate the model. The data underwent pre-processing with feature normalization to standardize input values and advance model convergence. The SWO, inspired by humpback whales’ bubble-net feeding approach, is employed to optimize the hyperparameters of the GAT, allow the technique to improved learn discriminative features from complex, graph-structured medical data. Experimental outcomes show that the SWO-GAN structure constantly outperforms conventional diagnostic techniques across all datasets, achieve notable improvement in accuracy values of Hepatitis (0.94211), Liver (0.98645), Parkinson’s (0.94123). This research highlights the potential of integrating bio-inspired optimization methods with graph-based deep learning (DL) to create scalable, precise, and clever diagnostic tools, contributing appreciably to the future of sustainable smart healthcare systems.