Abstract:The more training data there is, the slower training the support vector regression technique gives rise to. Thus, it is good to remove the non-support vector examples beforehand. In addition, small variation in kernel parameter has more influence on the generalization performance of the support vector regression approach than that in other parameters such as the regularization parameter C and the tube width ϵ. Therefore, kernel parameter optimization contributes to the small generalization error. This paper proposed a novel conformal kernel parameter transformation as well as training patterns reduction method to solve the regression problem. It determined the kernel center and kernel width in the factor function of kernel function based on data distribution of the high dimensional feature space. In particular, it preserved the volume around the boundary space, and reduced the spatial space on the non-support vector region. Moreover, it converted the regression problem to a classification problem, extracted the boundary points and then reduced the training patterns. Simulation results indicated that the present algorithm supported Housing dataset, improved the prediction accuracy by 10.76 percent over the primal approach, and reduced the number of support vectors from 192.4 to 144.7 over support vector regression. The proposed method also estimated the propylene concentration of propylene rectification tower. The experimental result of the propylene content demonstrated that our method decreased the prediction error by 14.07 percent over the neural network technique.

Regression Estimation Combined with Conformal Mapping and Boundary Extracting and It's Application

Fuzzy Support Vector Regression Model of 4-CBA Concentration for Industrial PTA Oxidation Process

Fuzzy Support Vector Machine For Regression Estimation

Support Vector Regression Based on Multi-Scale Wavelet Kernel for Propylene Concentration Estimation and Application

Conformal Transformation of Kernel Functions: A Data-Dependent Way to Improve Support Vector Machine Classifiers

Optimized Kernel-Based Conformal Predictor for Online Fault Detection

Virtual Sensors of Nonlinear Industrial Processes Based on Neighborhood Preserving Regression Model

Adaptive local weighted kernel-based regression for online modeling of batch processes

Soft Sensors of Nonlinear Industrial Processes Based on Self-Learning Kernel Regression Model

Regression with Support Vector Machines and VGG Neural Networks.

Nonlinear Regression with High-Dimensional Space Mapping for Blood Component Spectral Quantitative Analysis

Improved Robust Kernel Subspace for Object-Based Registration and Change Detection

Virtual Sensor Development for Multioutput Nonlinear Processes Based on Bilinear Neighborhood Preserving Regression Model with Localized Construction.

Modifications of Kernels to Improve Support Vector Machine Classifiers

Cutting Plane Method for Continuously Constrained Kernel-Based Regression

Improving Support Vector Machine Classifiers By Modifying Kernel Functions

Scaling the kernel function to improve performance of the support vector machine

Estimation of Convergence Rate for Multi-Regression Learning Algorithm

Conformalized Kernel Ridge Regression

Multi-scale support vector machine for regression estimation

Conformal Transformation of Kernel Functions