Model Efficiency Comparison for Heart Disease, Diabetes, and Parkinson's Prediction

Vijay U. Rathod's 2026 study proposes a unified deep learning framework evaluated on three benchmark clinical datasets. The framework integrates MLP, CNN, FT-Transformer, and autoencoder architectures, achieving AUC scores up to 0.8980 for heart disease prediction. Experimental results demonstrate competitive performance against classical machine learning baselines across heterogeneous biomedical data.

Use Cases

• Benchmarking deep learning model efficiency based on dataset characteristics like feature dimensionality and sample size
• Developing interpretable multi-disease prediction frameworks based on the modular architecture described
• Evaluating ensemble strategies (e.g., FT-Transformer + autoencoder) for clinical classification tasks based on the reported AUC metrics

Strengths

• Framework evaluated on three established benchmark datasets: UCI Heart Disease (303 samples), PIMA Indians Diabetes (768 samples), and Parkinson's disease voice (195 recordings)
• Reports specific model performance metrics, such as an AUC of 0.8980 (±0.0483) for heart disease prediction using an FT-Transformer + autoencoder ensemble
• Employs robust evaluation methods including fold-safe feature selection and nested cross-validation as described

Limitations

• Dataset is very small at 5.5 KB; the actual data files likely contain only summary results or model comparisons, not the full clinical records
• Column-level documentation is absent; field semantics must be inferred after download
• Row count for the comparison data is unknown, which may limit suitability assessment

Provenance

Source

figshare, author Vijay U. Rathod

Collection Method

Likely contains results from a computational study comparing deep learning frameworks on public clinical datasets.

Freshness

Last updated 2026-05-08 17:34:37