Support vector machines for system identification

Model selection approaches for non-linear system identification: a review

“…Some possible issues associated with SVR application in system identification have been discussed (Drezet and Harrison 1998). Although the SVR has the properties of sparsity control using "-insensitive loss function, it was found that the final SVR model may still have a large size despite the use of "-insensitive loss (Drezet and Harrison 1998;Lee and Billings 2002).…”

“…Although the SVR has the properties of sparsity control using "-insensitive loss function, it was found that the final SVR model may still have a large size despite the use of "-insensitive loss (Drezet and Harrison 1998;Lee and Billings 2002). There is current research into improving the sparsity of the SVM (Burges 1996;Downs, Gates, and Masters 2001).…”

“…Along with the growing popularity of SVM, successful applications have been reported in engineering system modelling and control (Drezet and Harrison 1998;Iplikci 2006), such as modelling of proton exchange membrane fuel cell (Zhong, Zhu, and Cao 2006) and quality monitoring of a plastic injection moulding process (Ribeiro 2005), etc. It has also been widely used in bioinformatics (Bradford and Westhead 2005;Tothill et al 2005), Geo-and environmental sciences (Kanevski et al 2004;Khan and Coulibaly 2006), and many other areas (Hong 2006), though one of the major recent contributions is on bioinformatics for classification of high throughput 'nomics' datasets, e.g.…”

Prediction of Thermal Performance of Designed Different Obstacles on Absorber Plates in Solar Air Collectors by Support Vector Machine

“…In the constrained minimization, kernels corresponding to data points that are within the error bounds are removed. The support vector regression (SVR) is formed by the retained kernels [18], and the data points associated with the retained kernels are referred to as the support vectors (SVs). Since the kernels of the SVR are similar to the basis functions of the radial basis function (RBF) network with scatter partitioning, it is shown here that the SVR can be reformulated as a RBF network with basis functions normalized such that they form a partition of unity [19].…”

A Big Data Analysis System for Financial Trading

“…The theoretical basis of SVM is the structural risk minimization principle, which gives excellent generalization properties. However, it has been shown that the standard SVM technique is not always able to construct parsimonious models in system identification (Drezet and Harrison, 1998). This shortcoming encourages the exploration of new methods for the parsimonious models under the framework of both SVM and KM.…”