Publisher's Synopsis
Modeling of any system or a process is one of the significant but challenging tasks in engineering. The challenge is either due to the physical complexity of natural phenomenon or our limited knowledge of mathematics. Recently, data driven modeling (DDM) has been found to be a very powerful tool in helping to overcome those challenges, by presenting opportunities to build basic models from the observed patterns as well as accelerating the response of decision makers in facing real world problems. Since DDM is able to map causal factors and consequent outcomes from the observed patterns (experimental data), without deep knowledge of the complex physical process, these modeling techniques are becoming popular among engineers. Soft computing and statistical models are the two commonly employed data-driven models for predictive modeling. As far as the statistical data-driven models are concerned, these models could be employed in the life of modern engineering. But the accuracy and generalizability of these models is very poor. The soft computing data- driven modeling techniques have attracted the attention of many researchers across the globe to overcome the limitations of statistical methods. The statistical data-driven modeling techniques such as least-squares methods, the maximum likelihood methods and traditional artificial neural network (ANN) are based on empirical risk minimization (ERM) principle while the support vector machine (SVM) method is based on the structural risk minimization (SRM) principle. According to it, the generalization accuracy is optimized over the empirical error and the flatness of the regression function or the capacity of SVM. On the other hand, the ANN and other traditional regression models which are based on ERM principle minimize the empirical error and do not consider the capacity of the learning machines. This results in model over fitting i.e. high prediction accuracy for the training data set and low for the test (unseen) data, giving poor generalization performance. SVMs belong to the supervised machine learning theory and are applied to both nonlinear classification called support vector classification (SVC) and regression or SVR. SVM possesses many advantages over traditional neural networks.
