Integration of Logistic Regression and Genetic Programming to Model Coastal Louisiana Land Loss Using Remote Sensing

Created by W.Langdon from gp-bibliography.bib Revision:1.7630

@InProceedings{Momm:2007:ASPRS,

• author = "Henrique G. Momm and Joel S. Kuszmaul and Greg Easson",
• title = "Integration of Logistic Regression and Genetic Programming to Model Coastal Louisiana Land Loss Using Remote Sensing",
• booktitle = "Proceedings of the ASPRS 2007 Annual Conference",
• year = "2007",
• editor = "Gary Florence",
• address = "Tampa, Florida, USA",
• month = "7-11 " # may,
• organization = "American Society for Photogrammetry and Remote Sensing",
• keywords = "genetic algorithms, genetic programming, remote sensing, logistic regression",
• URL = "http://www.asprs.org/a/publications/proceedings/tampa2007/0044.pdf",
• size = "8 pages",
• abstract = "The land loss along the Louisiana Coast has been recognised as a growing problem. Efforts have been concentrated in the creation of a Decision Support System (DSS) to better address the problem in which the correct water delineation from remotely sensed data is a critical part of this project. Two different approaches have been evaluated in previous studies: logistic regression and genetic programming. Herein a third approach is proposed by combining genetic programming with logistic regression. This hybrid approach merges the ability of logistic regression to deal with dichotomous data and to provide quantitative results with the optimisation characteristic of genetic programming to search the entire hypothesis space for the ``most fit'' hypothesis. Genetic programming modifies (using an iterative trial and error process) logistic regression models formed by vegetation indices built from basic function blocks defined in the function set (arithmetic operations) and in the terminal set (vegetation indices and spectral bands). Each candidate model is refined with a stepwise backward elimination using the level of significance associated with Chi-square test of each term and then evaluated based on the fitness function which is defined by: the model's, Kappa statistics and the number of terms in the model. The final output is a two-class (water and non-water) classified image of the most fit model.",
• notes = "http://www.asprs.org/conference-archive/tampa2007/",

}

Genetic Programming entries for Henrique G Momm Joel S Kuszmaul Greg Easson

Citations