A Matrix-based Visual Analytics approach for the Analysis of Association Rules

Abstract

With a large amount of data, many data mining techniques have contributed to the extraction of useful information from the data. The generated information, if interpreted correctly results in a huge benefit for the user. Association rule mining is a popular data mining technique because it helps in finding hidden relationships between items in a database. The application of this data mining technique is not just limited to market data but also in health, entertainment, and census data to, name a few. With large data, the rules generated can be in thousands and needs to be assisted with proper visualization to attain the correct information that cannot be collected using textual rules. Text-based rules make it difficult for a user to analyze the results easily and get more information at the same time. Previously, a lot of efforts have been made to assist the visualization of these rules but there exists a gap in the proposed systems. In this work, we present ARMatrix, a system that provides an item-to-rule matrix-based visualization. The system comprises four modules, with the main interface providing the visualization to assist the rules with a view to compare selected rules based on different measures. The ordering options help to get a rearranged visualization based on placement or color. The freedom to find the subsets of the rules for exploration is provided using the filter section. To attain reusability and revisit a state later, the history section is provided to retrieve, delete or store the required state. The proposed system helps the general user with some knowledge of the association rules to navigate the rules and find the relevant information. The usability of the proposed technique to visualize and analyze the association rules is illustrated using two user scenarios and then confirmed from the feedback received by conducting a user test with 20 participants.