KEEL is a software tool developed to build and use different
Data Mining models. The main features of KEEL are:
It contains pre-processing algorithms: transformation, discretization,
instance and feature selection and so on.
It also contains a Knowledge Extraction Algorithms Library, supervised
and unsupervised, remarking the incorporation of multiple evolutionary learning
algorithms.
It has a statistical analysis library to analyze algorithms.
It contains an user-friendly interface, oriented to the analysis of algorithms.
We can distinguish four parts in the graphic environment:
Data Management: the Data Management part allows users to create new Datasets or to create some partitions from an existing one. Also, you can view information from Datasets of your own, with the only restriction that they must meet the Keel format. In addition, it is possible to edit and apply transformations to the existing Datasets.
Another important thing is that the Data Management part allows you to generate Datasets in the keel format from UCI files.
Experiments: the Experiments part has the objective of designing the desired experiments
using a graphical interface. Doubtless, this is the more innovative tool integrated
in this tool. The objective is to use available datasets and algorithms to generate
a directory structure with all the necessary files needed to run the designed
experiments in the local computer selected by the user. Now, you can forget
scripts and other parameter files that made arduous the design of an experiment,
and begin to use the new windows based interface.
With this program, yon only need to select the input data (datasets), the algorithms
you want to use and to make the opportune connections between them. Also it
is possible to concatenate methods, insert statistical tests, etc ...
One of the tasks that was more simplified is probably the configuration of the
parameters; everything can be done from a simple dialog without requirement
of external configuration files.
This part of KEEL has two main objectives: on the one hand, you can use
the software as a test and evaluation tool during the development of an algorithm.
On the other hand, it is also a good option in order to compare new developments
with standard algorithms already implemented and available in KEEL 2.0.
Educational: the teaching part has the main objective of designing the desired experiments
using a graphical interface and an on-line execution of those experiments, being posible to stop and resume them as you need. Also, you can see the results of those experiments into the environment.
However, this part has a reduced number of available methods and lacks of result method and statistical test
Modules: This part includes new modules extending the functionalities of the KEEL software tool:
Imbalanced learning: A module specially designed for generating experiments with imbalanced
data.
Non-Parametric Statistical Analysis: This module allows to easily analize the results of any experimental study, expresed in raw CSV format. To do so, it includes several
well-known non-parametric statistical tests, ready to use.