4607

Iowa State University researchers have developed a novel algorithm approach for computing causal interaction from time series data. The proposed approach will help determine which time series data will cause or influence which other time series data, given two or more-time series data.

The detection between cause and effect (causality) has long been a question in many scientific area. There are various theories developed so far, but they failed to give the causal structure in dynamic systems. Besides, none of these definitions is capable of differentiation of direct and indirect influence, and identification of links from the system.
Iowa State University researchers provided a new definition of information transfer between the states of a dynamic system, and the definition captures the intuitions of information transfer, like transfer asymmetry, zero transfer and information conservation. Furthermore, this measure clearly distinguish between direct and indirect influence.
The researchers has demonstrated that how this information measure can be used to characterize influence in real world networks. For example, analyzing time-series data of two different stocks it will be possible to determine which stock is causing or influencing the other stock i.e., the direction and degree of influence can be determined using our proposed approach. Similarly, in application involving health care and medicine discovery it will be possible to determine which genes among millions of genes in the gene regulatory network is most responsible for causing genetic disorder and hence disease.

• Captures the true causal structure in dynamic system

• Distinguish between direct and indirect influence

• Demonstration of the approach through stock prices analysis

This novel algorithm approach will find application in various filed where the primary goal is to infer causal interaction between two or more quantities, such as stock market, health care, medicine discovery and etc.