An Approach for Automatic Discovery of Rules Based on ECG Data Using Learning Classifier Systems

Abstract

Personalized medicine aims to understand the underlying relationships between the multitudes of factors affecting a patient's health and provide physicians with an evidence-based approach to customize the treatment based on patient-specific characteristics. Machine-learning techniques can examine available data and discover relationships and patterns that may not be explicitly expressed within the data. In this case, physicians can use this knowledge for hypothesis testing and conduct investigations into the possible conditions that affect the patients' health. The benefits of personalized medicine include improved patient satisfaction, reduced length of hospitalization, enhanced treatment outcomes, and increased overall efficiency of the health care system. In this paper, we present an approach based on Learning Classifier Systems (LCS) to automatically discover rules that can support medical decision-making in evaluating the patient's heart condition. LCS are considered adaptive rule-based systems that can evolve a set of classifiers called rules based on a learning component that assigns credit to existing rules and an evolutionary component that helps discover new ones. The proposed approach is based on the implementation of an accuracy-based LCS that has been modified to support rules learning for personalized medical decision-making. The experimental results in the case study section provide a proof of concept for rules learning based on ECG data to discover rules that can support physicians in the medical decision-making process.