|
Here you can download
KnowledgeMiner free, investigate a variety of
examples,
and find links
to more information on Self-Organization
and KnowledgeMiner.
|
KnowledgeMiner
Description
|
KnowledgeMiner is a new data mining tool that enables anyone
to use its unique form of modeling to quickly visualize new
possibilities. It is an Artificial Intelligence tool
designed to extract hidden knowledge from data easily. It
was built on the cybernetic principles of self-organization:
Learning a completely unknown relationship between output
and input of any given system in an evolutionary way from a
very simple organization to an optimally complex one. The
main advantages of this inductive approach are:
- Only minimal, uncertain a priori information about
the system is required. That means, even if you have no
experience in modeling, data analysis or designing a
neural network you will be able to model, analyse and
predict very complex relationships of nearly any kind of
system.
- A very fast and effective learning process for a
personal computer. That means, you can solve problems on
your desktop in a reasonable time which you may have
never thougth possible before.
- Modeling very short and noisy data samples. That
means, you can deal with a problem as is and don't have
to construct artificial conditions for your modeling
method to get it work.
- Output of an optimally complex model. This means, you
generally can be sure to get a model at the end of the
automated modeling process which can be expected not to
be overfitted. Overfitted models are not able to predict
inherent relationships between variables.
- Output of an analytical model as a transparent
explanation component. That means, you can evaluate the
analytical model to explain the obtained results
immediately after modeling.
KnowledgeMiner version 3 works on three advanced
inductive learning modeling algorithms:
GMDH-type Neural Networks
- This method creates parametric time series models,
static or dynamic input-output models and predictable
systems of equations. Up to 500 input variables could be
considered for model creation, whereby at least 6 data
samples are needed for each variable. The network
structure is not predefined. A generated linear model may
look like this:
Self-organizing fuzzy rule
induction
- Working much like GMDH-type Neural Networks, this
method generates fuzzy rules from fuzzy or boolean data.
Using fuzzy variables like negative, positive or medium,
the generated rules are composed of several AND, OR, NOT
operators, and they show natural language-like
descriptive power:
Analog Complexing
Analog Complexing is a multidimensional pattern
search method that can be used for predicting most fuzzy
objects. It self-selects several similar patterns
relative to a given reference pattern and then uses their
known continuations to form a prediction for the
reference pattern.
In KnowledgeMiner, data is stored and edited in a
spreadsheet. All models created in a document have graphical
and analytical representations, and they are stored in a
model base. In this way, they are easily accessible and
applicable for prediction, classification or diagnosis tasks
within the program.
The power and the advantages of KnowledgeMiner compared
with statistics as well as with traditional neural networks,
make it easy to use and rapidly applicable to a wide range
of real-world problems, and characterize it as the most
effective modeling and prediction tool available.
KnowledgeMiner's algorithms can be used for different data
mining tasks:
|
julian@scriptsoftware.com 