Multi-objective Symbolic Regression to Generate Data-driven, Non-fixed Structure and Intelligible Mortality...

Area of Science:

• Computational Biology
• Machine Learning
• Medical Informatics

Background:

• Symbolic Regression (SR) is a data-driven method using Genetic Programming for creating mathematical models.
• SR generates interpretable, arbitrarily complex linear and non-linear functions, unlike fixed-structure statistical techniques.
• Its interpretability and flexibility offer advantages over traditional machine learning algorithms.

Purpose of the Study:

• To evaluate Symbolic Regression as a binary classifier for predicting in-hospital or short-term mortality in COVID-19 patients.
• To explore the capabilities and constraints of a novel SR implementation in a clinical setting.
• To compare SR's performance against established statistical and machine learning methods for mortality prediction.

Main Methods:

• Implementation of Symbolic Regression (SR) as a binary classification model.
• Application of the SR model to a dataset of COVID-19 patients for mortality prediction.
• Comparative analysis of SR performance with other statistical and machine learning methodologies.

Main Results:

• Symbolic Regression demonstrated competitive classification performance in predicting COVID-19 patient mortality.
• The SR approach showed stability in managing unbalanced datasets, a common issue in clinical data.
• SR models provided intrinsically interpretable results, enhancing clinical understanding.

Conclusions:

• Symbolic Regression presents a viable and competitive alternative for modeling clinical phenomena like mortality prediction.
• The technique's interpretability, performance, and stability make it valuable for analyzing complex health data.
• SR offers a powerful tool for medical informatics, particularly in predicting outcomes for diseases like COVID-19.