Explainable AI for Chronic Kidney Disease Prediction in Medical IoT: Integrating GANs and Few-Shot Learning

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Public Health Research

Background:

• Chronic kidney disease (CKD) is a growing global health concern.
• Machine learning (ML) shows promise for CKD identification but lacks clinical transparency.
• Opaque ML models hinder adoption in real-world healthcare settings.

Purpose of the Study:

• To develop an explainable ML framework for accurate CKD prediction.
• To address missing data challenges in CKD datasets using generative adversarial networks (GANs).
• To evaluate few-shot learning techniques combined with explainable AI for CKD classification.

Main Methods:

• Utilized generative adversarial networks (GANs) for data imputation in CKD datasets.
• Employed few-shot learning methods: prototypical networks and model-agnostic meta-learning (MAML).
• Integrated explainable AI techniques (SHAP, LIME) for model interpretability and compared with traditional ML models (SVM, LR, DT, RF, VEL).

Main Results:

• Few-shot learning models with GANs significantly outperformed traditional ML methods in CKD prediction.
• Prototypical networks with GANs achieved 99.99% accuracy; MAML reached 99.92%.
• Prototypical networks demonstrated high performance metrics (F1-score, recall, precision, MCC) on raw data.

Conclusions:

• The proposed framework offers a reliable and trustworthy solution for CKD classification.
• This approach enhances smart medical applications within the Medical Internet of Things (MIoT) ecosystem.
• The study facilitates accurate CKD prediction, detection, and optimal medical decision-making.