A Novel Interval Integer Genetic Algorithm Used for Simultaneously Selecting Wavelengths and Pre-processing Methods

Abstract

Abstract Because the primary coding version for chromosome was adopted in previous genetic algorithm (GA), a large number of inexplicable individuals were created in this procedure and the priority of pre-processing methods could not be optimized when GA was used for simultaneously selecting wavelength and pre-processing treatments. To solve the problem, a novel interval integer genetic algorithm (NIGA) was presented in this work. After validating by a group of synthesizing data and two groups of real near spectra data, NIGA was better than anyone of the other usual methods such as partial least squares regression (PLSR), interval partial least squares regression (iPLS) and ant colony optimization genetic algorithm sample selection (ACOGASS). During its work, common pre-processing methods were classified into four categories, e.g. smoothing, derivative, correction and standardization. The parameters of these methods were directly encoded as NIGA chromosomes for auto-adjustment at maximum extent. On the other hand, partitioning full spectrum was used for reducing the random noise and computational complexity, extracting the available information to improve the ability of the model. Furthermore, this method can also work in different situations and demands with the other non-linear models.