A dual-optimized deep learning pipeline for Chronic kidney disease (CKD) screening achieved Accuracy = 0.973 ± 0.022 and AUC = 0.996 ± 0.006 on the UCI CKD dataset (25 attributes) using nested stratified 5-fold cross-validation, while narrowing inputs to a minimum of 7 clinically important features. The article describes an SSS-BGSA-EHO-DBNN approach that automates feature selection and model parameter tuning to produce an interpretable screening workflow.

Why It Matters To Your Practice

• Earlier CKD detection can change trajectories, but many ML CKD tools are brittle on small/unbalanced datasets and require manual tuning that limits real-world adoption.

• This framework targets “hands-off” model development (automated feature selection + hyperparameter optimization) to reduce clinician-facing opacity and implementation friction.

Clinical Implications

• Potential to support automated CKD screening/flagging from routine structured data, with performance reported at AUC ~0.996 in this benchmark dataset.

• Feature parsimony: the model identified a minimum of 7 key features, which could simplify data requirements compared with high-dimensional EHR extracts.

• Top-ranked signals included specific gravity, hypertension, packed cell volume, glucosuria, and blood urea—variables clinicians can sanity-check against physiology and local practice patterns.

Insights

• Method stack: spiral search strategy-based binary gravitational search algorithm (SSS-BGSA) for feature selection + elephant herding optimization (EHO) to tune/train a deep belief neural network (DBNN).

• Design goal: improve the exploration–exploitation balance (SSS-BGSA) and convergence/learning efficiency (EHO), reducing manual trial-and-error.

• Evaluation used nested stratified 5-fold CV, a stronger setup than a single split for estimating performance on limited datasets.

The Bottom Line

• On UCI CKD, a fully automated, dual-optimized DBNN pipeline delivered high discrimination (AUC 0.996) while surfacing a small set of clinically recognizable features—suggesting a path toward more interpretable AI-assisted CKD screening, pending external validation.