Coron is a domain and platform independent, multi-purposed data mining toolkit, which incorporates not only a rich collection of data mining algorithms, but also allows a number of auxiliary operations. To the best of our knowledge, a data mining toolkit designed specifically for itemset extraction and association rule generation like Coron does not exist elsewhere. Coron also provides support for preparing and filtering data, and for interpreting the extracted units of knowledge.
💡 Deep Analysis
📄 Full Content
The Coron System
Mehdi Kaytoue1, Florent Marcuola1, Amedeo Napoli1, Laszlo Szathmary2, and
Jean Villerd1
1 Laboratoire Lorrain de Recherche en Informatique et ses Applications (LORIA)
Campus Scientifique – BP 239 – 54506 Vandœuvre-lès-Nancy Cedex (France)
{kaytouem, marcuolf, napoli, villerd}@loria.fr
2 Département d’Informatique – Université du Québec à Montréal (UQAM)
C.P. 8888 – Succ. Centre-Ville, Montréal H3C 3P8 (Canada)
Szathmary.L@gmail.com
Abstract. Coron is a domain and platform independent, multi-purposed
data mining toolkit, which incorporates not only a rich collection of data
mining algorithms, but also allows a number of auxiliary operations. To
the best of our knowledge, a data mining toolkit designed specifically for
itemset extraction and association rule generation like Coron does not
exist elsewhere. Coron also provides support for preparing and filtering
data, and for interpreting the extracted units of knowledge.
Key words: knowledge discovery, data mining, itemset extraction, as-
sociation rules generation, rare item problem
1
System Overview
Born for a particular need in a cohort study [1], Coron is now a framework
of knowledge discovery in databases on its own, used in several application do-
mains, e.g. [4–6]. Intended to an educational and scientific usage, the Coron
system is articulated into several modules for preparing and mining binary data,
and filtering and interpreting the extracted units. Thus, from binary data (pos-
sibly obtained from a discretization procedure), Coron allows one to extract
itemsets (frequent, closed, generators, etc.) and then to generate association
rules (non-redundant, informative, etc.). Building concept lattices is also pos-
sible. The system includes many classical algorithms of the literature, but also
others that are specific to Coron [9–11]. The software is freely available at
http://coron.loria.fr. Mainly written in Java, Coron is compatible with
the Unix, Mac and Windows operating systems and is of command-line usage.
2
A Global Data Mining Methodology
The methodology was initially designed for mining biological cohorts, but it is
generalizable to any kind of database. It is important to notice that the whole
process is guided by an expert, who is a specialist of the domain related to
the database. His role may be crucial, especially for selecting the data and for
arXiv:1111.5690v1 [cs.DB] 24 Nov 2011
2
M. Kaytoue, F. Marcuola, A. Napoli, L. Szathmary and J. Villerd
interpreting the extracted units, in order to fully turn them into knowledge units.
In our case, the extracted knowledge units are mainly association rules. At the
present time, finding association rules is one of the most important tasks in data
mining. Association rules allow one to reveal “hidden” relationships in a dataset.
Finding association rules requires first the extraction of frequent itemsets.
The methodology consists of the following steps: Definition of the study
framework; Iterative step: data preparation and cleaning, pre-processing step,
processing step, post-processing step; Validation of the results and Generation
of new research hypotheses; Feedback on the experiment. The life-cycle of the
methodology is shown in Figure 1. Coron is designed to satisfy the present
methodology and offers all the tools that are necessary for its application in
a single platform.
Pre-processing. These modules propose several tools for manipulating and for-
matting large data. The data are described by binary tables in a simple text-file
format: some individuals in lines possess or not some properties in column. The
main possible operations are: (i) discretization of numerical data, (ii) conversion
of different file formats, (iii) creation of the complement of the binary table, and
(iv) other projection operations such as transposition of the table.
Fig. 1. Architecture of the Coron System
The Coron System
3
Data mining. Extracting itemsets and association rules is a very popular task
in data mining. Concept lattices are mathematical structures supported by a
rich and well established formalism, namely, Formal Concept Analysis [13]. A
concept lattice is represented by a diagram giving nice visualization of classes of
objects of a domain. Thus, the data mining modules of the Coron System offer
the following possibilities:
– Itemset extraction: frequent, closed, rare, generators, etc. This task is per-
formed by a large collection of algorithms based on different search strategies
(depth-first, level-wise, etc.).
– Association rules generation: frequent, rare, closed, informative, minimal
non-redundant, Duquenne-Guigues basis, etc. These rules are given with
a set of measures such as support, confidence, lift, conviction, etc.
– Concept lattice construction.
Post-processing. Extracted units from the data mining step may be very numer-
ous, and hide some units of higher interest. Thus, Coron proposes some filtering
operations that should be done in interaction with a domain expert. The ana-
lyst may fi
