Temporal and Causal Relations on Evidence Theory: an Application on Adverse Drug Reactions
Keywords:Dempster-Shafer Theory, linear regression, time series, adverse drug reactions, information fusion, machine learning, natural language processing
The use of big data and information fusion in electronichealth records (EHR) allowed the identification of adversedrug reactions(ADR) through the integration of heteroge-neous sources such as clinical notes (CN), medication pre-scriptions, and pathological examinations. This heterogene-ity of data sources entails the need to address redundancy,conflict, and uncertainty caused by the high dimensionalitypresent in EHR. The use of multisensor information fusion(MSIF) presents an ideal scenario to deal with uncertainty,especially when adding resources of the theory of evidence,also called Dempster–Shafer Theory (DST). In that scenariothere is a challenge which is to specify the attribution of be-lief through the mass function, from the datasets, named basicprobability assignment (BPA). The objective of the presentwork is to create a form of BPA generation using analy-sis of data regarding causal and time relationships betweensources, entities and sensors, not only through correlation, butby causal inference.