Table 1.
Summary of literature on disease prediction models.
Fig 1.
Multi-model disease prediction framework.
Fig 2.
Abstract representation of the model families considered in this study.
Table 2.
Model configuration for heart disease dataset.
Table 3.
Model configuration for diabetes disease dataset.
Table 4.
Model configuration for Parkinson’s disease dataset.
Table 5.
(a) Standalone model training configuration (MLP, CNN, FT-transformer). (b) Ensemble model training configuration (MLP + AE, CNN + AE, FT + AE).
Table 6.
Standalone model search spaces.
Table 7.
Ensemble model hyperparameter grid.
Table 8.
Autoencoder architecture configuration.
Table 9.
Cross-validation configuration per dataset.
Table 10.
Summary of proposed instantiations.
Table 11.
Baseline model configurations.
Table 12.
The different evaluation metrics.
Table 13.
Heart disease classification performance.
Fig 3.
All models comparison.
Table 14.
Diabetes classification performance.
Fig 4.
The shows grouped performance metrics for diabetes classification.
Table 15.
Parkinson’s disease classification performance.
Fig 5.
The presents grouped performance metrics for Parkinson’s disease classification.
Table 16.
Classical ML baseline performance.
Table 17.
DL vs. classical ML – Best model comparison.
Fig 6.
Confusion matrix for FT + AE on heart disease.
Fig 7.
Confusion matrix for CNN + AE on diabetes.
Fig 8.
Confusion matrix for MLP on Parkinsons disease.
Fig 9.
ROC curve for CNN on heart disease.
Fig 10.
ROC curve for CNN on diabetes.
Fig 11.
ROC curve for MLP on Parkinson’s disease.
Fig 12.
Precision-recall curve for CNN on heart disease.
Fig 13.
Precision-recall curve for CNN on diabetes.
Fig 14.
Precision-recall curve for MLP on Parkinson’s disease.
Table 18.
The accuracy comparison with literature benchmarks.
Fig 15.
The compares reported accuracy of the proposed models with benchmark results from prior studies.
Table 19.
Selected Wilcoxon signed-rank test results (AUC).
Table 20.
Model efficiency comparison.
Table 21.
Parameter scaling across datasets.
Fig 16.
SHAP summary plot for FT on heart disease (best fold).
Fig 17.
FT-Transformer CLS attention to features: Heart disease.
Fig 18.
Feature cross-attention heat map: Heart disease.
Fig 19.
Mean grad-CAM activation map for CNN on diabetes.
Fig 20.
Decision curve analysis for FT + AE on heart disease.
Fig 21.
Decision curve analysis for CNN on diabetes.
Fig 22.
Decision curve analysis for MLP on Parkinson’s disease.
Table 22.
PPV/NPV across prevalence rates (Best standalone model per dataset by AUC, best fold).
Table 23.
Ablation study: Incremental impact on mean AUC.
Table 24.
Error distribution summary (Best model per dataset by AUC, best fold).