Discovery method and system suitable for a plurality of hidden confounding factor data

Abstract

The invention discloses a causal network discovery method and system suitable for a plurality of hidden confounding factor data. The system comprises a user input data module, a parameter configuration module, a causal skeleton learning module, a local causal sequence identification module, a hidden confounding factor detection module, a causal network construction module and a result visualization module. The causal network discovery system suitable for the plurality of hidden confounding factor data can discover the causal network according to the needs of users, and then visually present the causal network between observation variables. The invention takes into account the case of the plurality of hidden confounding factors, the information of causal skeleton can greatly reduce the complexity of solution, and detect the number of implicit confounding factors and the observed variables of their effects, so that the learning and identification accuracy of the causal network are high,and the problem that the causal network can not be found correctly in the case of multiple implicit confounding factors data is solved. The invention is simple in implementation and high in learning accuracy, and has strong practical value and practical significance.

Description

technical field [0001] The invention relates to the technical field of machine learning, in particular to a method and system for discovering a causal network applicable to multiple hidden confounding factor data. Background technique [0002] With the advent of the era of big data, in many fields, discovering the causal relationship between things can help people discover potential information from a large amount of data and understand the real mechanism behind things. However, how to effectively mine causality from data and apply it to various fields of our lives has been a hot topic of research. Although the linear non-Gaussian acyclic model (LiNGAM) proposed by Shimizu et al. is widely used because it is easy to understand and estimate the causal relationship from the data, but in reality, most data contain hidden confounding factors (that is, unobservable variables ), discovering causal structure from these data poses a challenge for us. [0003] Although most studies...

AI Technical Summary

Problems solved by technology

However, the complexity of this algorithm is very high, and multiple regressions and (conditional) independence tests are required. At the same time, the number of hidden confounding factors cannot be detected, let alone the relationship between hidden confounding factors and observed variables.

[0005] Some of the existing methods for applying the LiNGAM model to causal structure discovery cannot be applied to the presence of multiple confounding factors, and cannot effectively identify hidden confounding factors and discover the causal relationship between them and observed variables, so a new method is proposed. A method and system that is close to the data, low in complexity, effectively detects the number of hidden confounding factors, and discovers the causal relationship between observed variables and hidden confounding factors has certain research value and significance

Images