Parameter Scaling Across Datasets: A Unified Framework for Multi-Disease Prediction

A study by Vijay U. Rathod proposes a dataset-aware deep learning framework for disease prediction, evaluated on three benchmark clinical datasets. The framework integrates multiple architectures like MLP, CNN, and FT-Transformer, using robust preprocessing and nested cross-validation. Results include an AUC of 0.8980 for heart disease prediction and 0.8451 for diabetes classification on the tested datasets.

Use Cases

• Benchmarking deep learning models for clinical prediction based on the described framework and datasets
• Developing interpretable disease classifiers based on the modular architecture that selects models by dataset characteristics
• Evaluating ensemble strategies like FT-Transformer + autoencoder for heart disease prediction based on the reported AUC performance
• Studying model performance across heterogeneous data types based on the included voice, metabolic, and clinical feature datasets

Strengths

• Framework evaluated on three distinct benchmark datasets: UCI Heart Disease (303 samples), PIMA Diabetes (768 samples), and Parkinson's voice (195 recordings)
• Reports specific performance metrics, such as an AUC of 0.8980 (±0.0483) for heart disease prediction
• Employs robust methodological practices including fold-safe feature selection and nested cross-validation

Limitations

• Row count is unknown, which may limit suitability assessment
• Column-level documentation is absent; field semantics must be inferred after download
• The dataset is very small at 5.5 KB, indicating limited scope, likely containing framework parameters or summary results rather than raw clinical data

Provenance

Source

Vijay U. Rathod via figshare

Collection Method

Likely contains parameters or results from the described machine learning study.

Freshness

Last updated 2026-05-08 17:34:38; freshness should be verified