Kmap Clustering Demo

Suppose you just returned from 2 weeks vacation and are sorting the mail that's accumulated into piles -- bills, personal mail, magazines, junk mail, etc. In some ways, that's what a Kohonen neural network clustering program does, with one big plus -- it arranges the piles into a two-dimensional array in such a way that two piles next to each other are more "alike" than two piles that are farther apart. You can see and interact with some results yourself as soon as the green "GO" buttons appear.

NEW: Prescription Pattern Demo

Here, Kmap displays Neuralware clustering results on the prescription patterns of a set of doctors. The actual names of the drugs have been substituted by the condition they are designed to treat. Note that the header of the topten drilldown table now provides a breakdown by specialty of the doctors who are prescribing the drugs in the chosen segment.

Supermarket Demo

A minute or so after you hit the green GO button you should see a 6x6 grid showing the top 3 items purchased by each of 36 segments of customers. In each segment, the darkness of the outside color indicates population, while the inside color indicates mean spending. Right-click on a segment and the popup menu gives you access to two very powerful drilldowns. You can also choose 7 different sort keys from "Sort By" on the menubar.

Credit Line Usage Demo

We applied IBM Intelligent Miner's Kohonen clustering method to data on the fraction of total credit line used by 1000 households over a 12-month period. The fraction stayed pretty constant at about 30% for the population as a whole (top graph), but the 2x2 Kohonen map in the middle shows the population has at least 4 different segments, whose outstanding balances are (0) constantly high, (1) falling, (2) rising, and (3) constantly low. The color of each square indicates its population; the exact number is available via a right mouse click. The 4x4 map (bottom) shows even more different patterns, but the Kohonen network has pulled a little trick on us: the rows and columns are flipped with respect to the 2x2 map, which makes comparison difficult. No problem! Pull down the View menu, select flip rows & columns, and voila! We could continue on up to more segments, but the point is made: this population has numerous patterns of credit usage, and knowing this can be invaluable in tailoring a marketing strategy.

Stock Price Behavior Demo

Here we show some of the "drill-down" capability of the Kmap visualizer. One year's end-of-month closing prices for the S&P 500 stocks were divided into 64 segments using Intelligent Miner's neural clusterer. If you right-click on the square for node 0 , you'll see that the popup menu now lists the stocks in that segment. Exxon and Shell are both there. How similar were they? Right-click to get the popup menu and select DrillDown to see the individual behavior of the stocks in this segment. How did Texaco do by comparison? Pull down View, click Find what segment ..., select Texaco from the list, and push Find: Segment 32 should turn yellow. Right-click on segment 32, and you'll see Mobil in there too. You may have other questions. Have fun exploring!

The Kmap visualizer used in this demo was created by George Almasi using the JCChart Java Beans package from Sitraka.