The Analysis of Association Rules: Sensitivity Analysis

Abstract

Association rules are extracting information from transactional databases of documents with a collection of items also called “tokens” or “words”. The approach is used in the analysis of text records, of social media and of consumer behaviour. We present an innovative sensitivity analysis of association rules (AR) measures of interest. In text analytics, a document term matrix (DTM) consists of rows referring to documents and columns corresponding to items. In binary weights, “1” indicates the presence of a term in a document and “0” otherwise. From a DTM one computes measures of interest characterising ARs. The approach we introduce is based on the application of befitting cross validation (BCV) principles to ARs. The sensitivity analysis of ARs is based on computer generated repeated shuffling of training and validation sets that provide an assessment of the uncertainty of AR measures of interest. We demonstrate this methodology with reports of symptoms associated with a Nicardipine drug product used in the treatment of high blood pressure and angina. Patients self-reports on side effect events are analysed. Association rules, derived from these reports, describe combinations of terms in these reports. AR measures of interest are defined in section 1. In section 2 we introduce the case study that motivates the method we propose. In section 3 we apply BCV principles by concatenating side effect events of Nicardipine by patient. Sensitivity analysis (SA) of ARs is introduced and demonstrated in section 4. The sensitivity analysis method presented here is discussed in section 5 where we formulate general data analysis considerations on how to organise and analyse semantic data.