Symbolically Regressing Fish Biomass Spectral Data: A Linear Genetic Programming Method With Tunable Primitives.

Machine learning techniques play an important role in analyzing spectral data. Fish biomass spectral data is useful in fish production as it carries key chemical properties of fish meat. However, it is challenging for existing machine learning techniques to comprehensively discover hidden patterns from fish biomass spectral data because the spectral data often have considerable noise, while the training data are quite limited. To better analyze fish biomass spectral data, we modeled it as a symbolic regression problem and solved it by a linear genetic programming method with newly proposed tunable primitives. In the symbolic regression problem, linear genetic programming automatically synthesized regression models based on the given primitives and training data. The tunable primitives further improved the approximation ability of the regression models by tuning their inherent coefficients. Our empirical results from over ten fish biomass targets showed that our proposed method using tunable primitives improved the overall performance of fish biomass composition prediction. The synthesized regression models are compact and have good interpretability, allowing us to highlight useful features over the spectrum. Our further investigation verified the good generality of the proposed method across various spectral data treatments and other symbolic regression problems.