New Challenges in Association Rule Mining: an Approach Based on Genetic Programming

Created by W.Langdon from gp-bibliography.bib Revision:1.3963

@PhdThesis{Luna-JM:thesis,
  author =       "Jose Maria Luna Ariza",
  title =        "New Challenges in Association Rule Mining: an Approach
                 Based on Genetic Programming",
  school =       "Department of Computer Science and Artificial
                 Intelligence, University of Granada",
  year =         "2014",
  type =         "Ph.D. in Computer Science",
  address =      "Spain",
  month =        jan,
  keywords =     "genetic algorithms, genetic programming, association
                 rule mining",
  URL =          "http://www.uco.es/grupos/kdis/docs/thesis/2014-JMLuna.pdf",
  size =         "224 pages",
  abstract =     "This Doctoral Thesis involves a series of approaches
                 for mining association rules by means of a
                 grammar-guided genetic programming based methodology.
                 The ultimate goal is to provide new algorithms that
                 mine association rules in only one step and in a highly
                 efficient way. The use of grammars enables the
                 exibility of the extracted knowledge to be increased.
                 Grammars also enable obtaining association rules that
                 comprise categorical, quantitative, positive and
                 negative attributes to be mined.

                 Firstly, as for the mining of frequent association
                 rules, a novel grammar-based algorithm, called G3PARM,
                 has been proposed. It is able to discover rules having
                 positive, negative, categorical and quantitative
                 attributes. The evolutionary model is able to perform
                 the mining process in one single step. This PhD Thesis
                 also includes a model for mining rare or infrequent
                 association rules, as well as two multi-objective
                 approaches that optimise two different quality measures
                 at time. Additionally, two novel algorithms that
                 self-adapt their parameters are considered. In this
                 sense, a previous tuning of the parameters would not be
                 required, as they are adjusted depending on the data
                 under study. Finally, the developed methodologies have
                 been applied to the educational field to discover
                 interesting information that could be used to improve
                 the courses.

                 All the algorithms proposed in this Doctoral Thesis
                 have been evaluated in a proper experimental framework,
                 using different types of datasets and comparing their
                 performance against other published methods of proved
                 quality. Results have been verified by applying
                 non-parametric statistical tests, demonstrating the
                 many benefits of using a grammar-based methodology to
                 address the association rule mining problem",
  notes =        "In English. Supervisors Sebastian Ventura and Jose
                 Raul Romero",
}

Genetic Programming entries for Jose Maria Luna

Citations