Subscribe Bookmark RSS Feed
bernard_mckeown

Joined:

Jun 23, 2011

Data Mining, Modeling and Visualisation at the British Museum

I walked it into the British Museum one morning last week, eager to begin this special day for JMP. An Explorers seminar would take place here soon, and we would show delegates from a range of industries the latest incarnation of JMP: JMP Pro. And this, with the blessing of one of the leading statistical thinkers, Professor Richard De Veaux.


Professor De Veaux, or Dick as his friends know him, proved to be great company, not just for the day, but also for the evening before. Dick and I joined Jeff Perkinson, JMP Product Manager, for dinner at Gordon Ramsay’s Maze restaurant, where great food was tasted and great conversation was had. After an evening of dissecting and discussion, we came to realise the following:


1. Data mining is about advising.

2. There are two kinds of advising where data mining is concerned.

3. The first is traditional data mining, where a department (for example, Arrears) is provided with a list to target based on scoring code. Many tools, including JMP Pro, can do this.

4. The second advising is about king making. These are the “right-hand persons” to senior execs, who advise them on how to make the right decisions. JMP Pro is uniquely placed to do this because of its amalgamation of visualisation with advanced modelling. In other words, with JMP Pro, you can show the exec what you think rather than simply telling them.


The day itself was fantastic, a great learning environment for the many delegates from banking, manufacturing, marketing communications and R&D who attended the event. Dick positioned data mining as an investigative approach and outlined his seven steps to successful data mining:


1. Define the goal or problem.

2. Bring the data into a single, accessible environment.

3. Explore the data to get an idea of what is important within it.

4. Prepare the data for modelling – steps 1-4 will typically take 60-95% of your time!

5. Build a model of the data.

6. Evaluate that model.

7. Deploy it – either as scoring code to produce lists, or for executive decision making.


In the afternoon, we showed how JMP Pro addresses the key issues highlighted in the Rexer Report 2010, a survey of data miners:


1. Dirty data and data accessibility are two of the top three issues highlighted by data miners. Robert Anderson showed some simple ways in which you can address both of these.


2. The ability to visualise and explain models is highlighted consistently as problem with many software packages. Malcolm Moore and Jeff Perkinson showed how JMP Pro can be used to combine advanced modelling with world-class visualisation to make results accessible – even to management.


The Rexer Report highlights that on average a data miner has access to “4.6” tools (though how you can have access to 0.6 of a tool is beyond me!). Dick explained this with an analogy about surgeons: If someone was operating on you, would you want him or her to have access to the best and most advanced tools possible, if and when the surgeon needed them? Of course, you would! Well, the same should be the case for your company, as its health could depend on having access to the best tools -- which is why your company should have JMP Pro.

Article Tags