3/30/2012

Predictive Analysis

Predictive Analysis.

What is it?

Look for patterns over time to guesstimate future behavior.

When I worked in the Mortgage business for Nations Bank in 1995, they purchased a new software system which calculated if a customer was credit worthy.

They based the customer's credit score, including charge offs, slow pays, foreclosures, etc. and bounced that info off the customer's length of residence, length of employment and number of recent inquiries.

So based off a limited number of factors, they could predict whether a customer was going to fulfill the length of the loan or if the customer would default.

Quite fascinating.

A fortune teller, for profit.

 We were instructed to only override the system a certain percentage of times, for extenuating circumstances.

My job was to compile the approvals and declines on a daily basis.

That's actually how I entered IT, because I convinced the bank to pay for a college class at the St. Petersburg College - I took C++ and got the highest grade in the class, and the professor spoke with a friend of his and placed me at a part time position doing Visual Basic 4.

Anyhow, the basis of predicting future behavior based on previous action has been around for a while.

However, because data and big data have become so big recently, along with data mining, predictive analysis can be utilized to process larger data sets in quicker time, by the average BI person.

So for example, if my company sold sporting goods equipment, and I gathered weather patterns for the past 100 years, and I noticed a steady increase in snow in the NE USA, I could then stock up on gloves and scarves.  However, I could also bounce that data against the average annual income of the people in the NE, to see if disposable income has increased.  If so I could promote the latest high end ski equipment and make record profits.

Now take the opposite, sun screen.  If the weather patterns indicated increased warmth, obviously people would be stocking up on sun block.  To confirm this, you could mash the data with medical data such as skin cancer treatments over time to see if there was a medical necessity.  Thus you could clean up on that too.

Both those examples are weather based, but they are only examples, and you can substitute numerous others.

You could look at baseball batting percentages, stadium seating capacity / tickets sold, average age of hall of fame players, best coaching statistics, anything really.

This is all well and good, assuming the people crunching the numbers are looking to profit and better humanity in some way or educate people or medical purposes.

Overall Predictive Analysis can't tell you who will win the World Series for the next 5 years and what the scores will be in game 7 five years from now.

There is only so much predicting one can do, as outside influences / forces skew the future behavior.

A pitcher gets injured, a team goes on a hitting streak, a team brings up 3 minor league players who rock the majors.

You can study the past as much as you want, but some things can't be predicted.

You would have to know all factors at the time of decision, which is nearly impossible.

So we have to make decisions from the data after careful analysis, looking for patterns over time, blended with intuition and business experience, to make the best possible decision at the time.

That's my take on it anyway.

No comments:

Post a Comment

Note: Only a member of this blog may post a comment.

Thoughts to Ponder