Statistical causality analysis

Abstract

The problem of identifying deterministic cause-and-effect relationships, initially hidden in accumulated empirical data, is discussed. Statistical methods were used to identify such relationships. A simple mathematical model of cause-and-effect relationships is proposed, in the framework of which several models of causal dependencies in data are described - for the simplest relationship between cause and effect, for many effects of one cause, as well as for chains of cause-and-effect relationships (so-called transitive causes). Estimates are formulated that allow using the de Moivre-Laplace theorem to determine the parameters of causal dependencies linking events in a polynomial scheme trials. The statements about the unambiguous identification of causeand-effect dependencies that are reconstructed from accumulated data are proved. The possibilities of using such data analysis schemes in medical diagnostics and cybersecurity tasks are discussed.

Keywords

• finite classification task
• cause-and-effect relationships
• machine learning