A Genetic Programming-Driven Data Fitting Method

Abstract

Data fitting is the process of constructing a curve, or a set of mathematical functions, that has the best fit to a series of data points. Different with constructing a fitting model from same type of function, such as the polynomial model, we notice that a hybrid fitting model with multiple types of function may have a better fitting result. Moreover, this also shows better interpretability. However, a perfect smooth hybrid fitting model depends on a reasonable combination of multiple functions and a set of effective parameters. That is a high-dimensional multi-objective optimization problem. This paper proposes a novel data fitting model construction approach. In this approach, the model is expressed by an improved tree coding expression and constructed through an evolution search process driven by the genetic programming. In order to verify the validity of generated hybrid fitting model, 6 prediction problems are chosen for experiment studies. The experimental results show that the proposed method is superior to 7 typical methods in terms of the prediction accuracy and interpretability. INDEX TERMS Data fitting, hybrid model, genetic programming, tree coding, interpretability.

Two-Stream Deep Fusion Network Based on VAE and CNN for Synthetic Aperture Radar Target Recognition

“…At present, several probabilistic statistical models have been developed to describe HRRP [23,[26][27][28]. Nevertheless, traditional probabilistic models need to preset the distribution patterns of data, such as Gaussian distribution and Gamma distribution, which are relatively simple and have some limitations in their data fitting ability (the ability of fitting the original data distribution) [29]. In addition, since traditional probabilistic models are based on shallow architectures with simple linear mapping structures, they are only good at learning linear features.…”

Microscopic Characterizations of Heterogeneous Pores, ITZs, Multiple‐Solids, and Their Impacts on Damage Property of Sandstone by Low‐High Resolution 3D Reconstruction

“…To find the most reliable statistical model to characterize the microscopic heterogeneity of sandstone rocks, the trial‐and‐error method (Chen et al., 2020; Yang et al., 2017; Z. Zhao & Zhou et al., 2020a) is applied, in which the Lorentz relation, Logistic relation and SineSqr relation are selected to initially represent the mechanical parameter distribution of the ITZs, pore and grain phases consisting of microscopic elements, as shown in Figure . By comparing these relations, the optimization statistical model to represent the mechanical property distribution of microscopic elements for sandstone is determined.…”