Bankruptcy Prediction Demo
As part of a larger datamining study (described here
and here) we transformed about
1 million Credit Bureau records of households with marginal credit
scores (650 to 680) by filtering out many of the 200+ items per
record and by expanding each household's revolving credit balance
into 12 categories, based on the account's age and credit
limit. A data mining program then divided these households
into 36 groups using these new categories.
link brings up a separate display* of these groups in a rectangular grid, as shown below;
Colored by Score...
...and by Subsequent Bankruptcies
On the left, the color of each
rectangle reflects the average credit rating of each
group, pink being lower and blue being higher. The
top 3 credit categories of each group are also shown.
Not much of a pattern is visible when the rectangles
are colored by their credit score.
But click the "Color By" menu in the demo and
select "Bankruptcies" and you'll get the coloration shown
on the right above. Now the color of each rectangle indicates the
fraction of households in that group that went bankrupt in the following
year, and a pattern does appear, showing that the datamining program
clustered the higher bankruptcy risks into the lower right corner
of the grid. In group 30, for example, the bankruptcy rate is 3.8%,
which is 13 times higher than average. Note that many of the other
groups in this marginal-credit pool in fact have lower-than-average
|Now click the "BB Signatures" menu
to see the pattern of the 12 categories we described above,
and note the characteristic "bankruptcy signature"
visible for group 30. Here's how an analyst interprets group
- The squares marked "1" indicate they once were able
to get credit lines over $7500, but not in the last two
- The squares marked "2" show they have a "sweet spot" in credit lines between $2000 and $7500 ... easy to get and they quickly add up tp a lot of money.
- The squares marked "3" show they have been unable to get new credit in the last 6 months as their "risk" score elevates.
Customers and lenders both benefit when a finer filter
like this is used to separate the credit-worthy from those
who pose a higher risk.
*If the Demo button fails, try disabling "Ad Blocking" in
your firewall, i.e. Norton Internet Security; also, you
can see what you missed