Search

Twin support vector regression (TSVR) aims at finding 휖-insensitive up- and down-bound functions for the training points by solving a pair of smaller-sized quadratic programming problems (QPPs) rather than a single large one as in the conventional SVR. So TSVR works faster than SVR in theory. However, TSVR gives equal emphasis to the points above the up-bound and below the down-bound, which leads to the same influences on the regression function. In fact, points in different positions have different effects on the regressor. Inspired by it, we propose an asymmetric ν-twin support vector regression based on pinball loss function (Asy- ν-TSVR). The new algorithm can effectively control the fitting error by tuning the parameters ν and p. Therefore, it enhances the generalization ability. Moreover, we study the distribution of samples and give the upper bounds for the samples locating in different positions. Numerical experiments on one artificial dataset, eleven benchmark datasets and a real wheat dataset demonstrate the validity of our proposed algorithm. [ABSTRACT FROM AUTHOR]

Support vector regression (SVR) and neural network (NN) models were used to predict average daily gain (ADG), feed efficiency (FE), and feed intake (FI) of broiler chickens during the starter period. Input variables for construction of the models were levels of dietary protein and branched-chain amino acids (BCAA; valine, isoleucine, and leucine). Starting with 241 lines, the SVR and NN models were trained using 120 data lines and the remainder (n = 121) was used as the testing set. The SVR models were developed using different kernel functions including: linear, polynomial (second and third order), radial basis function (RBF), and sigmoidal. In order to evaluate the SVR models, their performance was compared to that of multilayer perceptron (MLP)- and RBF-type NN models. Results indicated that MLP-type NN models were the most accurate in predicting the investigated output variables (R2 for ADG in training and testing = 0.81 and 0.81; FE = 0.87 and 0.87; FI = 0.68 and 0.62). Among the different SVR kernels, best performance was achieved with the RBF (R2 for ADG in training and testing = 0.76 and 0.76; FE = 0.85 and 0.87; FI = 0.46 and 0.48) and polynomial (third order) function (R2 for ADG in training and testing = 0.77 and 0.77; FE = 0.85 and 0.87; FI = 0.46 and 0.39), whose prediction ability was better than that of the RBF-type NN (R2 for ADG in training and testing = 0.75 and 0.75; FE = 0.82 and 0.82; FI = 0.41 and 0.39) models. Sigmoidal SVR models provided the poorest prediction. The work demonstrates that, through proper selection of kernel functions and corresponding parameters, SVR models can be considered as an alternative to NN models in predicting the response of broiler chickens to protein and BCAA. This type of model should also be applicable in poultry and other areas of animal nutrition. [ABSTRACT FROM AUTHOR]

In this paper, a simplification of the necessary and sufficient Karush-Kuhn-Tucker (KKT) optimality conditions for the Lagrangian support vector regression in 2-norm resulting a naïve root finding problem of a system of equations in m variables is proposed, where m is the number of training examples. However, since the resulting system of equations contains terms having the nonsmooth “plus” function, two approaches are considered: (i) the problem is reformulated into an equivalent absolute value equation problem and solved by functional iterative and generalized Newton methods and (ii) solve the original root finding problem by generalized Newton method. The proofs of convergence and pseudo-codes of the iterative methods are given. Numerical experiments performed on a number of synthetic and real-world benchmark datasets showing similar generalization performance with much faster learning speed in comparison with support vector regression (SVR) and as good as performance in comparison with the least squares SVR, unconstrained Lagrangian SVR proposed in Balasundaram et al. (Neural Netw 51:67-79, 2014), twin SVR and extreme learning machine clearly indicate the effectiveness and suitability of the proposed problem formulation solved by functional iterative and generalized Newton methods. [ABSTRACT FROM AUTHOR]

Accurately, forecasting of the flatness plays a highly significant role in the flatness theory and flatness control system, but it is quite difficult and complicated due to the nonlinear characteristics of flatness pattern recognition and lack of available observed data set. Recently, support vector regression (SVR) is being proved an effective machine learning technique for solving nonlinear regression problem with small sample set, because of its nonlinear mapping capabilities. However, it has also been proved that the prediction precision of SVR is highly dependent of SVR parameters, which are hardly choosing for the SVR. As in many excellent algorithms, gravitational search algorithm (GSA) not only has strong global searching capability, but also is very easy to implement. In the paper, an improved gravitational search algorithm (IGSA) is presented to further enhance optimal performance of GSA, and it is employed to serve as a method for pre-selecting SVR parameters. Based on SVR and IGSA algorithms, a forecasting model of flatness pattern recognition is proposed. Where, the IGSA is employed to optimize the parameters of SVR model to determine the parameters as fast and accurate as possible. Afterward, a procedure of forecasting flatness was put forward to evaluate the efficiency of the proposed IGSA-SVR model, which was compared with normal SVR model, IGSA-BP model and extreme learning machine model. The results affirm that the proposed algorithm outperforms other technique. [ABSTRACT FROM AUTHOR]