Premium
Ant colony optimisation: a powerful tool for wavelength selection
Author(s) -
Shamsipur Mojtaba,
ZareShahabadi Vali,
Hemmateenejad Bahram,
Akhond Morteza
Publication year - 2006
Publication title -
journal of chemometrics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.47
H-Index - 92
eISSN - 1099-128X
pISSN - 0886-9383
DOI - 10.1002/cem.1002
Subject(s) - ant colony optimization algorithms , overfitting , selection (genetic algorithm) , computer science , heuristic , mathematical optimization , ant colony , population , path (computing) , algorithm , artificial intelligence , mathematics , demography , sociology , artificial neural network , programming language
Ant colony optimisation (ACO) is a meta‐heuristic algorithm, which is derived from the observation of real ants. Real ant colonies are distributed system that, in spite of the simplicity of their individuals, present a highly structured social organisation and can accomplish complex tasks. They always find a short path between the nest and a food source. ACO is based on local message exchange via the deposition of pheromone trails. It is in fact a population‐based approach using positive feedback as well as greedy search. Wavelength selection is a strategy used for improving the quality of calibration methods. As a first report, this work indicated that the ACO possesses a great ability to find best subsets of wavelengths, at a short period of time with small PRESS values, via accumulation of information in the form of pheromone trails deposited on each wavelength. Theory of ACO is described and, to carry out the wavelength selection, a fitness function is defined. The ACO parameters are configured with a 3‐levels full factorial design. The high ability of ACO in wavelength selection process was demonstrated by examining four different NIR and UV/Vis data sets via various ACO algorithms, including ACO‐ILS, ACO‐CLS and ACO‐PLS. The results showed that, with the same fitness function, ACO‐ILS algorithm suffers from some overfitting problem. This problem was overcome by constraining the algorithm to choose limited number of wavelengths, the corresponding algorithm called as ACO‐ILS(limited). The results obtained by these algorithms clearly revealed the improved predictive ability of ACO in wavelength selection over the existing full‐spectrum models. Copyright © 2007 John Wiley & Sons, Ltd.