- author = "Ben McKay and Mark J. Willis and Geoffrey W. Barton",
- title = "Using a Tree Structured Genetic Algorithm to Perform Symbolic Regression",
- booktitle = "First International Conference on Genetic Algorithms in Engineering Systems: Innovations and Applications, GALESIA",
- year = "1995",
- editor = "A. M. S. Zalzala",
- volume = "414",
- pages = "487--492",
- address = "Sheffield, UK",
- publisher_address = "London, UK",
- month = "12-14 " # sep,
- publisher = "IEE",
- keywords = "genetic algorithms, genetic programming",
- ISBN = "0-85296-650-4",
-
DOI = "
doi:10.1049/cp:19951096",
- size = "6 pages",
- abstract = "In this contribution a tree structured genetic algorithm is described. The algorithm is used to generate non-linear models from process input-output data. Three examples are used to demonstrate the applicability of the technique within the domain of process engineering.",
-
notes = "12--14 September 1995, Halifax Hall, University of
Sheffield, UK see also
http://www.iee.org.uk/LSboard/Conf/program/galprog.htm
Uses correlation coefficient in fitness function as advocated by M. C. South Phd 1994 'The application of GAs to rule finding in data analysis', Newcastle upon Tyne, UK
Final fixup? 'Mutation...replaces a node in the tree with another of the same degree'. Elitist. Pop size 20, G=100, Pcross=0.8 Pmut=0.5 'found to give good performance to date'
'non linear least-squares optimization to obtain 'best' value of the (new) constant(s) in the expression'. 'the fitness of a tree is weighted according to its size' (penalise bigger) Anti-bloat
2nd example 'Near Infra-red reflectance instrument for the inference of the protein contents of ground wheat' (old data, (1983, T.Fearn 'A misuse of ridge regression in the calibration of near infrared reflectance instrument', Appl Statistics, 32, 1, 73-79), various techniques already tried). GP 'provide simple non-linear model that provides far greater insight into the input-output model structure than other non-linear modelling techniques such as neural networks' RMS error also better than cited in literature (traditional stats and ANN).
3rd: recovery of contaminated transformer oil. GP solution robust to measurement error.",
