Data mining can be described as data processing using sophisticated data search capabilities and statistical algorithms to discover patterns and correlations in large pre-existing databases (Agrawal & Srikant 1995; Zhao & Sourav 2003). From these patterns, new and important information can be obtained that will lead to the discovery of new meanings which can then be translated into enhancements in many current fields. In this paper, we focus on the usability of sequential data mining algorithms. Based on a conducted user study, many of these algorithms are difficult to comprehend. Our goal is to make an interface that acts as a “tutor” to help the users understand better how data mining works. We consider two of the algorithms more commonly used by our students for discovering sequential patterns, namely the AprioriAll and the PrefixSpan algorithms. We hope to generate some educational value, such that the tool could be used as a teaching aid for comprehending data mining algorithms. We concentrated our effort to develop the user interface to be easy to use by naïve end users with minimum computer literacy; the interface is intended to be used by beginners. This will help in having a wider audience and users for the developed tool.
TopBackground
Kopanakis and Theodoulidis (2003) highlight the importance of visual data mining and how pictorial representation of data mining outcomes are more meaningful than plain statistics, especially for non-technical users. They suggest many modeling techniques pertaining to association rules, relevance analysis, and classification. With regards to association rules they suggest using grid and bar representations for visualizing not only the raw data but also support, confidence, association rules, and evolution of time.
Eureka! is a visual knowledge discovery tool that specializes in two dimensional (2D) modeling of clustered data for extracting interesting patterns from them (Manco, Pizzuti & Talia 2004). VidaMine is a general purpose tool that provides three visual data mining modeling environments to its user: (a) the meta-query environment allows users through the use of “hooks” and “chains” to specify relationships between the datasets provided as input; (b) the association rule environment allows users to create association rules by dragging and dropping items into both the IF and THEN baskets; and (c) the clustering environment for selecting data clusters and their attributes (Kimani, et al., 2004). After the model derivation phase, the user can perform analysis and visualize the results.
TopMain Thrust
AprioriAll is a equential data pattern discovery algorithm. It involves a sequence of five phases that work together to uncover sequential data patterns in large datasets. The first three phases, Sorting, L-itemset, and Transformation, take the original database and prepare the information for AprioriAll. The Sorting phase begins by grouping the information, for example a list of customer transactions, into groups of sequences with customer ID as a primary key. The L-itemset phase then scans the sorted database to obtain length one itemsets according to a predetermined minimum support value. These length one itemsets are then mapped to integer value, which will make generating larger candidate patterns much easier. In the Transformation phase, the sorted database is then updated to use the mapped values from the previous phase. If an item in the original sequence does not meet minimum support, it is removed in this phase, as only the parts of the customer sequences that include items found in the length one itemsets can be represented.
After preprocessing the data, AprioriAll efficiently determines sequential patterns in the Sequence phase. Length K sequences are used to generate length K+1 candidate sequences until K+1 sequences can no longer be generated (i.e., K+1, is greater than the largest sequence in the transformed database. Finally, the Maximal Phase prunes down this list of candidates by removing any sequential patterns that are contained within a larger sequential pattern.