IEEE Geoscience and Remote Sensing Magazine - December 2015 - 31

contemporaneously, in the same software environment,
a multitude of tools (e.g., Matlab, R, Python, Perl scripts
and for instance RapidMiner or Weka clustering tools)
that can be used within a workflow allowing a user/developer to take advantage of the benefits derived from each
computational language, in the same Java environment.
One example is R, which is known for the numerous
statistical packages and the ability of developing sophisticated data plots [5]. KNIME has specific R nodes that
can be implemented internally to process data and plot
graphics using R.
The KNIME processing pipeline works in steps with
nodes upstream being evaluated first, and sequentially
downstream nodes being evaluated progressively. Each
node supports, depending on its internal configuration,
different data visualizations (tabular and more complex
statistical visualizations such a scatter-plots, contingency tables, etc.). The advantage is that each workflow
is not a "black box" where input/output are the only
"accessible parts of the analysis" rather the analysis flow
and relative data manipulation can be monitored progressively in each node, facilitating significantly debugging sessions and making the analysis and data processing more transparent to the user. In some cases the nodes
will represent a specific statistical model rather than a
relatively simple data-processing step, such as treating
Missing Information (MI) in the data. For these cases,
more complex visualizations are available to facilitate
the interpretation of modelling results (e.g., decision
trees, k-means, SOMs).
KNIME supports workflow abstraction, since any developed workflow or part of a workflow can be condensed into
a so-called Meta-Node. This process is also referred to as
encapsulation and offers the opportunity to the programmer to focus on higher level programming. Another advantage of meta-nodes is that they can be copied and nested to
facilitate iterations of the same workflow over different data
sets in the same workflow canvas, which facilitates data
integration. This concept has been recently complemented
by the loop concept from version 2.0 of the KNIME platform, which allows to process multiple data files iteratively
from the same workflow [4].
3. THE KNIME DEVELOPER/USER COMMUNITY
An active community of KNIME users and developers
supports constant software upgrades, which is one of the
key advantages of free open source software. This results
in extensive libraries of nodes that can facilitate the solution of disparate problems, including earth science related
problems. The availability of a Learning Hub (http://www.
knime.org/learning-hub?src=knimeapp) is also an important source of information for beginners. The KNIME
Analytics Platform is available for free under the GNUv3
Public License for Windows, Linux and Mac OS X (10.7 and
above) at the following web address: http://www.knime.
org/downloads/overview with also a Developer Kit (SDK)
december 2015

ieee Geoscience and remote sensing magazine

option that simplifies node development in the Java runtime environment.
4. KNIME FUNCTIONALITY
Available KNIME nodes cover a vast range of functionalities; nodes can be categorized as follows:
◗ I/O: retrieves data from files or databases, export data in
a variety of formats.
◗ Data Manipulation: pre-processes the input data with
filtering, group-by, pivoting, binning, normalization,
aggregation, joining, sampling, partitioning, and other
operations such as replacement of missing information.
◗ Views: visualizes data and results through several interactive views, allowing for interactive data exploration.
◗ Mining: uses state of the art data mining algorithms
like clustering, rule induction, decision tree, association
rules, naïve Bayes, neural networks, and support vector
machines to name a few, allowing to perform common
pattern recognition, dimensionality reduction and feature reconstruction tasks. These are all areas of machine
learning that have seen significant application in the last
10 years in the geosciences, see [6], [7], [8].
Some of the most widely used programming languages
(Java, Python, R, Octave, and Matlab) are found in the
KNIME library as snippet nodes.
Beside the proprietary KNIME
nodes, programmable nodes
provide almost unlimited flexKNIME ALLOwS
ibility. These nodes can be comFLExIbLE DATA
bined in hybrid coding soluPROCESSINg AND
tions, taking advantage from
PATTERN RECOgNITION
the different armoury of tools
ANALYSIS IN A
available in each software lanMULTI-PLATFORM
guage/platform. Other nodes are
defined as wrappers, which inteENVIRONMENT.
grate functionality from third
party libraries. In particular,
KNIME integrates functionality
of several open source projects
that essentially cover all major areas of data analysis, one
example is Weka [9] and JFreeChart for visualization [1].
5. APPLICATIONS OF KNIME
The vast array of Machine Learning and Data Mining (DM)
tools in KNIME make the tool applicability very wide.
KNIME is presently used in numerous fields (e.g., Environmental Science, Customer Intelligence, Finance, Manufacturing, Pharma/Health Care, Retail, Text and Network
Mining) although as mentioned most case studies are in the
cheminformatics and bioinformatics areas. For instance,
[10] define KNIME as a non-domain specific solution and
integration backbone, with strong data pre-processing and
data analytics capabilities. The authors show how KNIME
can be extended and used in bioinformatics; in particular
for next-generation gene sequencing. Gene classification
involves extremely large data repositories and the authors
31

http://www http://www.knime.org/learning-hub?src=knimeapp http://www.knime

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - December 2015