A modular deep learning architecture for interpretable disease prediction across tabular clinical and biometric datasets

Clinical Snapshot

Key Findings

• P Value: Not reported

• Effect Size: AUC 0.8980 (heart disease), 0.8451 (diabetes), 0.7538 (Parkinson's)

• Primary Outcome: Disease prediction accuracy across three clinical conditions

• Nnt Or Sensitivity: Perfect specificity achieved for Parkinson's disease detection with MLP model

• Confidence Interval: ±0.0483 (heart disease), ±0.0270 (diabetes), not reported (Parkinson's)

Clinical Application

Moderate - requires computational infrastructure and technical expertise for implementation Limited applicability without validation on Australian populations; would require TGA approval for clinical decision support use; potential integration with existing electronic health records Patients requiring screening or early detection for heart disease, diabetes, or Parkinson's disease in clinical settings

Abstract

Accurate disease prediction using clinical datasets is essential for improving early diagnosis and clinical decision-support systems; however, many existing deep learning approaches are disease-specific, computationally intensive, and difficult to generalize across heterogeneous biomedical datasets. This study addresses this challenge by proposing a unified and dataset-aware deep learning framework that enables accurate and interpretable disease prediction across diverse clinical datasets. The framework adopts a modular architecture that selects appropriate models based on dataset characteristics such as feature dimensionality, sample size, and class imbalance. It integrates multiple deep learning architectures, including MLP, one-dimensional CNN, FT-Transformer, autoencoder-based classifiers, and ensemble strategies. Robust preprocessing, fold-safe feature selection, and nested cross-validation are incorporated to ensure reliable performance evaluation. The framework is evaluated on three heterogeneous benchmark datasets: the UCI Heart Disease dataset (303 samples, 13 clinical features), the PIMA Indians Diabetes dataset (768 samples, 8 metabolic features), and the Parkinson's disease voice dataset (195 recordings, 22 acoustic features). Experimental results demonstrate competitive predictive performance relative to classical baselines across the diverse tasks. The FT-Transformer + autoencoder ensemble achieved an AUC of 0.8980 (±0.0483) for heart disease prediction, while the CNN + Autoencoder ensemble obtained an AUC of 0.8451 (±0.0270) for diabetes classification. For Parkinson's disease detection, the MLP achieved an AUC of 0.7538 with perfect specificity. Overall, all models achieved AUC values comparable to ML baselines. The study contributes a scalable and interpretable deep learning framework that improves reliability, generalization, and practical applicability for multi-disease prediction in real-world healthcare environments.

References

1. 1.Rathod, V. U., Amrutkar, S. S., Patil, K. A., Londhe, A. D., Bobade, S. Y., Dhotre, V. A., & Kebede, M. W. (2026). A modular deep learning architecture for interpretable disease prediction across tabular clinical and biometric datasets. PLoS One. https://doi.org/10.1016/j.tbench.2025.100219