Estimation of Kolmogorov-Chaitin complexity on continuous biomedical data for machine learning purposes

Abstract

In the field of scientific research, the quest for understanding and explaining complex phenomena often involves uncovering patterns and underlying structures. To achieve this, researchers employ various methodologies and approaches. One such approach gaining increasing attention is the utilization of the Kolmogorov-Chaitin complexity (KCC). As a fundamental concept in information theory, KCC measures the information required to describe a particular object or data set. While initially defined for discrete data, recent advances have extended the concept to continuous data. This article aims to explore the estimation of Kolmogorov-Chaitin complexity on continuous biomedical data. It uses two publicly available data sets of large dimensions: the first consists of more than 400 h of EEG recordings, and the second includes over 40000 cases of sepsis and non-sepsis cases from the intensive care unit, in both cases as multivariate/multimodal time series. The estimation of KCC is a crucial procedure applied to these data. The results are subsequently used for machine learning (ML) purposes with the goal of prediction/classification. The ML models’ performance using this approach exceeds 0.95 by the area under the ROC curve.