Continuous Optimization of a Hierarchical Bayesian Network for Friedreich's Ataxia Severity Classification.

Abstract

Machine learning algorithms for rare disorders, such as Friedreich's Ataxia (FRDA), often suffer from a lack of data. Therefore, the ability for continuous optimization of an objective assessment model would be very useful as a clinical decision support system. In this study, we propose a Bayesian Network(BN) system for FRDA severity estimation that incorporates a Bayesian Statistical updating system to continuously improve the predictive ability while providing an easily interpretable graphical model. This can work to improve the understanding of the model by the clinician, thus creating trust in the machine learning process. Furthermore, we demonstrate that by using the updating mechanism, the BN model gives a goodness-of-fit score of 0.95, a root mean square error of 9.35 and a mean absolute error of 6.72, which outperforms other regression approaches as well as improves upon the base BN by 2% in goodness of fit, roughly 1% in RMSE and 6% in MAE.