Improved Visualization of Frequent Itemset Relationships Using the Minimal Spanning Tree Algorithm

Descriptive data mining techniques offer a way of extracting useful information out of large datasets and presenting it in an interpretable fashion to be used as a basis for future decisions. Since users interpret information most easily through visual means, techniques which produce concise, visually attractive results are usually preferred. We define a method, which converts transactional data into tree-like data structures, which depict important relationships between items contained in this data. The new approach we propose is offering a way to mitigate the loss of information present in previously developed algorithms, which use mined frequent itemsets and construct tree structures. We transfer the problem to the domain of graph theory and through minimal spanning tree construction achieve more informative visualizations. We highlight the new approach with comparison to previous ones by applying it on a real-life datasets - one connected to market basket data and the other from the educational domain.