Back to publications

Beyond Accuracy: Enhancing Parkinson’s Diagnosis with Uncertainty Quantification of Machine Learning Models

Asif Azad 1,3, Md. Saiful Islam 1,2, Ehsan Hoque 2,3, M. Saifur Rahman 1

  1. 1Bangladesh University of Engineering and Technology, Dhaka, Bangladesh
  2. 2University of Rochester, Rochester, NY, USA
  3. 3Atria Health and Research Institute, New York, NY, USA
AIiH 2025 · Springer LNCS August 2025
First author AIiH 2025 Best Paper Award nominee

Abstract

As machine-learning systems show growing promise for clinical diagnosis, establishing their reliability is essential for responsible medical deployment. This work evaluates Monte Carlo Dropout, Deep Evidential Classification, and Bayesian Neural Networks across motor, facial, and speech datasets for Parkinson’s disease detection.

Deep Evidential Classification performs poorly in both diagnostic accuracy and uncertainty assessment, while Monte Carlo Dropout and Bayesian Neural Networks provide more dependable uncertainty estimates. Identifying ambiguous predictions through uncertainty estimation can reduce diagnostic errors and support safer adoption of AI in medicine.

Citation

Azad, A., Islam, M. S., Hoque, E., & Rahman, M. S. (2025). Beyond Accuracy: Enhancing Parkinson’s Diagnosis with Uncertainty Quantification of Machine Learning Models. In Artificial Intelligence in Healthcare (AIiH 2025), LNCS 16038, 33–46. Springer.