Data-mining
Softopt has developed a data-mining package which uses
a new, unpublished algorithm to analyze data and use it
for making forecasts and predictions. It is a traditional tree-based
classifier system, but the separations are found with hybrid genetic
algorithms
which result in high-quality separations.
In comparison to neural networks our algorithm has the advantages
of delivering forecasts in the form of probabilities with a statistical
significance, of constructing forecast models much faster, of being
able to describe characteristic subgroups and of avoiding data-overfitting.
In comparison to traditional
decision trees it has the advantages of discovering correlations
between attributes much better, of avoiding the problem of data-overfitting
and of delivering forecasts in the form of probabilities with a statistical
significance.
Some example applications for the Softopt-data-mining package:
- An insurance company has a large data-base of policy holders from the
past.
Each past policy holder can be judged to have been belonging to a certain
risk-group. The insurance company wants to use personal information
about a policy-holder (like age, income, marital status .....) to
predict the probability that he belongs to the various risk-groups.
The quality of such predictions is basically what defines the
profit that a company makes.
- In medicine it is often desirable to predict whether a patient has
or does not have a disease from easily obtainable information about the
patient like blood-pressure, sex, past illnesses ....
However, one wants to keep the error-rate of misjudging an ill
patient to be healthy very low, but one tolerates a larger
error-rate of misjudging a healthy patient to be ill.
There exists a large data-base of patients from the past containing the
easily obtainable information and the information whether the
patients had or did not have the disease.
- Advertising agencies often have large records of past advertising
campaigns and the reaction the campaign got from people.
They want to use the information from the past to design a more specific
campaign and approach the right kind of people.
SoftOpt offers consulting for data-mining problems.
We offer to solve your data-mining problem,
part of the solution can be a software tailored to your problem.
The advantages for you are:
-
Fast solutions for your data-mining problems
SoftOpt has many advanced and well tested
data-mining algorithms implemented in software at its disposition.
We can choose the best algorithm suitable for
your specific problem.
-
There is no need for you to have an employee with
knowledge and skills in data-mining
Part of our solution for your problem is a software which
is tailored specifically for
your problem.
The user interface is easy and data input and output can be done
using comprehensible words. (The conversion and reconversion
in and from numbers respectively takes place in the software.)
Everybody can produce forecasts and analyses easily and confidently.
With the present data-mining packages on the market you need an
employee with data-mining skills who will spend weeks if not months
with the modeling of your data-mining problem.
-
Confidentiality of your data is guaranteed by the way we
collaborate with you
If you have confidential data then we come to you, install
the necessary software on one of your computers and work at your office.
It will not be necessary to transfer data from your computer
to one of ours at any time.
-
SoftOpt has developed some unique specialist data-mining algorithms
for particular application domains
For classification problems with many attributes (200 to 1000) SoftOpt
has developed a brand new decision tree algorithm which uses
genetic algorithms to find the separations and which decomposes
each separation problem in a tree-like manner in subproblems with
only very few attributes. This method is extremly fast considering
the size of these problems and the quality of the separations is
better than that of all the other conventional separation methods.
For another type of classification problem where the task is to
localize a small class SoftOpt has developed another very good
algorithm which outperforms the conventional data-mining methods.
-
Objektive comparisons between different methods and
algorithms are possible
For the objective evaluation of predictions of probabilities
SoftOpt has developed a method which is publicly available for free.
This evaluation method is contained in the software which is part
of a problem solution by SoftOpt and it enables you to objectively
compare various data-mining methods with respect to the
reliability of the forecasts.
-
Forecasts can be made very fast on every PC
even for large data sets
The data-mining process contains two parts:
- Construction of a forecast model, the result being a model file.
- New forecasts are produced using this model file.
Most of the computing power is always needed for step 1. for the
construction of the forecast model.
Step 2. is always very fast. Forecasts are produced in a fraction of
a second on any PC. Having millions of data of the past you can
produce the model file on a powerful mainframe computer, then
copy the model file on small PCs and laptops and produce
the forecasts there.