Visual Data Mining with Item Explorer

By Mihael Ankerst.

Basics, basics, basics – let’s get down to the basics:

With all the latest buzz about big data and the challenges in terms of scalability, one major aspect of data mining or analytics is not getting the attention it deserves: Where in terms of analytics is the division of labor between the human and the computer heading? The sometimes forgotten goal of data mining is to find patterns or relationships which are valid, new and applicable. Of these three criteria the algorithms are just addressing the first one, discovering patterns which are valid in some statistical sense.Deciding whether the discovered patterns are new, meaningful and applicable means to relate the new data driven insights to a business domain. This will remain the task of the humans for years to come since the steadily growing data available for the analytical task always covers just some limited aspects of a domain.Much knowledge is not accessible, not available in a form that can be used by current mining algorithms or simplyis just in the heads of smart people.

Visual data mining approaches try to address the challenge described above. Visualization can leverage the efficient pattern recognition capabilities of the user and combined with interactive options, the user can steer the exploration process by tying together the patterns obtained from the data with domain knowledge and ad-hoc hypotheses. The visual data mining perspective is to design a process which integrates automated computation with human exploration.

Item explorer is an open-source visual data mining tool based on the Javascript visualization library d3.js. Item explorer lets you do market basket analysis by exploring data interactively. Typically, market basket analysis is done by runningan algorithm to find association rules. However, the input parameters are hard to specify and even then the result set might still be very large.

Exploring frequent item sets with a moderate number of items can be done by representing the item frequencies with a bar chart. Each bar is representing one item frequency independent of other items potentially bought together. The bar chart (yes – going back to the basics!) turns out to be a very versatile representation. On the one hand, the cognitive ease and intuition to read and quantitatively interpret bar charts is still unmatched and on the other hand, meaningful interactions can extend the common static bar chart into a powerful tool to explore a search space with thousands or millions of item combinations.

Item explorer is freely available at:http://www.ankerst.de/Mihael/proj/mbc/

Summary: Item explorer is an open source visual data mining tool based on d3.js. It enables the user to interactively explore combinatorial questions such as analyzing frequent item sets.

Bio: Mihael Ankerst is a recognized expert in visual data mining. He is currently heading a data analytics group at Allianz.

Related

• Drag and Drop Visuals in your Interactive Dashboard – Gridster & D3.js
• Story-Boarding in a D3.js dashboard using Intro.js
• 21 Essential Data Visualization Tools