Encontrando regras de associação sem especificar suporte e confiança

Abstract

The extraction of information and knowledge in databases has been assuming a relevant role in aiding decision making. One of the main areas of research is association rule mining. This area makes it possible to capture relationships between attributes present in a database. Most algorithms used to extract association rules use support and confidence as parameters. Support represents the proportion of a given rule in the database and confidence represents the validity of this rule. Thus, professionals responsible for data analysis need to identify and define support and confidence thresholds (minimum support and minimum confidence, respectively) to obtain association rules. However, in certain contexts, it is difficult to identify good values for support and confidence in order to obtain the desired rules. In these situations, it may be necessary to run several queries with di↵erent values of support and confidence in order to obtain the desired rules. The purpose of this research is to examine association rules mining techniques and algorithms capable of obtaining association rules without the need to specify support and confidence, propose new algorithms and analyze these algorithms in terms of performance and quality of the obtained rules.