|
Seer Analytics
Enriching Marketing Data to Predict Customer Behavior
Seer Analytics, LLC is a research company with a unique
approach to compiling, understanding and presenting data.
Integrating survey research, database analysis, geographic
information systems and predictive modeling, Seer Analytics is in
the business of "transforming data into market insight." The
company works with organizations that generate very large amounts
of data, and enriches that data with various kinds of information.
Often that means developing predictive models to understand
customer behavior. For this, Seer Analytics has employed many
different modeling products.
Challenge: Who Should I Take to Lunch?
One client is a non-profit organization with millions of records.
The client needs to determine which members of the organization
are most likely to make a large donation (in the client's words,
"Who should I take to lunch?"). Certain types of analyses are
straightforward; for example, if a donor made a large donation in
the past, he or she is statistically a good candidate to make a
sizable gift in the future. However, many times it is not at all
obvious and Seer Analytics needs to examine interactions between
the various "features" (or variables) to find the best candidates.
Some relationships are quite subtle, and the patterns in the data
can be very difficult to tease out.
Solution: Models that Predict Behavior Using Non- linear Data
As the client's donor data was non-linear and fairly "noisy", Seer
Analytics decided that using neural networks technology would
provide the best results. NeuralWorks Predict from NeuralWare was
the tool chosen to perform predictive modeling on the data.
According to CEO Bill Lazarus, "In our business, which consists of
analyzing millions and millions of records for our clients
everyday, we use a wide variety of modeling tools. When it comes
to finding patterns in non-linear
Read the rest of this article... |
|
| |
Message from the CEO
Greetings,
With this issue of Synapse, we bring you a success story from
Bill Lazarus of Seer Analytics. Neural networks have always been
considered a core technology for market analysis - Bill's
commentary clearly demonstrates why.
On the product development front, we are well along with adding
some very sophisticated Self-Organizing Map capabilities to
NeuralSight, and the DuPont Acurum team has begun to use them as
it develops models for new markets in Asia and Europe. Alex Kulik
will be visiting Toronto in May to help ensure that the Acurum
team can take full advantage of the new features, as well as to
obtain all-important customer feedback that helps NeuralWare
continue to improve its products.
Look for more information about NeuralSight 2.0 in future Synapse
newsletters - and watch for the official release during the third
quarter of 2005! NeuralWare's other international activities also
continue apace.
By the time you read this, Mr. Kazuo Toda, Mr. Yukinori Tamura,
Mr. Toshiki Kashiwakura, and Mr. Yasuhiro Matsushita from our
partner in Japan, SET Software Ltd., will have arrived in
Pittsburgh for a day of project engineering discussions and sales
and marketing planning for the Japanese market. In a new joint
initiative, SET Software and NeuralWare will increase efforts to
offer NeuralWare technology to the growing embedded system and
robotics markets, since both the Pittsburgh region and the City of
Osaka, where one of SET Software's main offices is based, have
targeted Robotics Technology as a driver of future economic
growth.
Also, the paper I mentioned in the last issue of Synapse was
accepted for presentation at ICNC/FSKD'05, so I expect to be in
Changsha China in August. In addition, NeuralWare, Oracle/Retek,
and AbacusChina Ltd. submitted a joint proposal to Hunan
Television Station, based in Changsha, to design and implement the
world's most advanced television shopping channel for Hunan TV.
I hope you find this Synapse interesting - and if you would like
to have your work with NeuralWare technology featured in a future
Synapse, please contact me directly.
I would like Synapse to become increasingly customer-focused in
the coming months.
|
| |
|
| |
|
Tech Tips
|
| |
Neural Network Speed of Execution
By Bob Everly
Ever
wonder just how fast neural networks are at processing data? It
turns out that they are extremely fast - on the order of millions
of records per minute!
As an example we created a dataset of 100 input fields and 1
target output. A neural network was created in Professional II/PLUS,
trained to 50,000 learn iterations, tested on the same dataset,
and converted into C source code using our FlashCode feature. The
entire Professional II/PLUS session was completed in just under 2
minutes. This 100-input, 20-hidden layer, 1-output model is
reasonably large as most applications go.
The generated C code was then easily included into a main program
written specifically to measure speed of execution. The result:
10,000,000 records were processed in 76 seconds (1.3 minutes;
131,578 records per second).
The test machine is a 2.66GHz Pentium 4 computer with 768MB of
memory, although much less memory would have sufficed for this
test. Typical Windows XP background tasks were running such as
mail, virus checker, clock and so on.
The program presented a floating-point array of 100 elements to
our generated subroutine, which returned a predicted value into
another floating-point array. Nothing was done with the predicted
value such as writing it to the screen or disk, and no data was
read or posted into the 100-element input data array - this test
measured only the neural network through-put times. |
| |
Read the rest of this tech tip... |
| |
|
Have Your Modeling Story Published in Synapse
|
| |
If you would like to have your modeling story published in
Synapse, please send an email to
newsletter @neuralware.com
with a brief note about your application.
|
|
| |
|
The
professional's choice for predictive analytics
|
| |
With all the different tools out there, there is a reason
why NeuralWare is the professional's choice, used by most
Fortune 500 corporations and in companies and universities
all over the world. Call (412) 278-6292 or email
John Wavle
today to learn more about how we can help you with your
predictive modeling initiatives.
Learn More... |
|
| |
|
