The impact of threshold parameters in transactional data analysis

Today's information systems store large quantities of transactional data which may contain valuable but hidden information. Association rules generation is a method developed for analysis of this type of data. This method is however very resource-intensive. Both the execution time and the final model highly depend on threshold parameters set by the analysts. In this paper we analyze the impact of the minimal support parameter on the number of closed frequent itemsets discovered for various datasets. The analysis is conducted on referential datasets as well as real-life datasets which classify transactional elements in categories organized in hierarchical manner. Datasets that are not originally transactional can be transformed into a transactional form. They exhibit somewhat different relations between the minimal support parameter and the number of discovered closed frequent itemsets. Findings presented in this paper can serve as guidance in setting up support as the most important parameter affecting the final execution time. In order to analyze data characteristics from another perspective, we also present how varying confidence, lift and support can affect the number of formed association rules containing two elements.