China
Africa
Explore chapters and articles related to this topic
Pattern Recognition with Support Vector Machines
Published in Samuel D. Stearns, Don R. Hush, ®, 2016
Samuel D. Stearns, Don R. Hush
The design and implementation of SVM classifiers is accomplished using the libsvm software package, which employs an SMO training algorithm51 very much like the one described in Section 12.4.3.3. The instructions for downloading and installing libsvm for MATLAB can be found at the website listed in reference number 49. The libsvm software implements several types of SVMs, but defaults to the so-called C-SVM, which is equivalent to the SVM developed in this chapter. More specifically, the C-SVM algorithm is designed to optimize Vapnik’s original criterion s^λ(w,b)=12||w||2+C∑i=1n[1−yi(wTϕ(xi)+b)]+
Implementation
Published in Seyedeh Leili Mirtaheri, Reza Shahbazian, Machine Learning Theory to Applications, 2022
Seyedeh Leili Mirtaheri, Reza Shahbazian
LibSVM is a comprehensive library for Support Vector Machines (SVMs), and its development began in 2000 at National Taiwan University. This library is written C/C++ and partly Java. Learning duties of LibSVM include distribution estimation, support vector classification (SVC) for binary multi-class, and support vector regression (SVR). It supports C-SVC, v-SVC, ε-SVR, and v-SVR and distribution estimation (one-class SVM) formulations.
Quantitative Analysis of Aeroengine Turbine Disk Surface Crack under Natural Magnetic Field
Published in Research in Nondestructive Evaluation, 2022
Ping Fu, Bo Hu, Weitao Luo, Shaofei Wang, Xiwang Lan
In weak magnetic detection, the relationship between magnetic anomaly and the defect size parameters is nonlinear and complex. Moreover, the acquired data are limited in practical engineering. SVM has good generalization ability for learning small sample data and avoids the problem of over fitting [36]. Therefore, SVM is suitable for quantitative analysis of weak magnetic detection. The basic idea of SVM is to transform the samples into high-dimensional space, search the classification hyperplane in high-dimensional space, and establish nonlinear mapping to realize the linear separability of data samples in low-dimensional space. A library for support vector machines (LIBSVM) is a widely used SVM algorithm. The essence of the quantitative problem of weak magnetic detection is to establish the relationship between magnetic anomaly and defect size parameters.
Support vector machine classification applied to the parametric design of centrifugal pumps
Published in Engineering Optimization, 2018
E. Riccietti, J. Bellucci, M. Checcucci, M. Marconcini, A. Arnone
A choice that deeply influences the performance of an SVM learning algorithm is that of the kernel function (10). In LIBSVM, many different choices are possible: linear, polynomial, radial basis function or sigmoid. For this specific application, the best performance was obtained with the radial basis function kernel: where x, y are features and γ is a parameter to be set. A good choice of the free parameter γ is crucial. LIBSVM provides a tool to select it by cross validation on a set of values. However, when the dataset is large, cross validation can be a really time consuming process, so γ was determined using the default method employed in Tsang, Kwok, and Cheung (2005) and Ñanculef, Frandi, Sartori and Allende et al. (2014), that is, it was set to the average squared distance among training patterns.
Cross-wavelet aided ECG beat classification using LIBSVM
Published in Computer Methods in Biomechanics and Biomedical Engineering: Imaging & Visualization, 2018
Priyadarshiny Dhar, Saibal Dutta, Prithwiraj Das, Vivekananda Mukherjee
This paper addresses analysis of ECG signal based on cross-wavelet transform and LIBSVM classifier. Using LIBSVM no external tools are required for finding parameters as it is required in other algorithms such as genetic algorithm. LIBSVM is more advantageous because it provides an automatic model selection tool for C-SVC. In this paper, total 97,461 heartbeats are identified into normal and arrhythmia beats using 1000 training beats. Less than 1% of total heartbeats are used to train the classifier. The proposed binary classification scheme has been developed by utilising a small training data-set and a vast testing data-set to demonstrate the generalisation capability of the scheme. Such smaller training data-set decreases the training time and increases the accuracy as compared to other neural networks. The total classification accuracy has been found to be 96.66%, which is exceptionally high compared to NEWPNN, BPNN and ERNN classifier. Moreover, LIBSVM can also identify more abnormal beats than the other classifiers. Thus, the proposed algorithm provides an efficient, simple and fast method for determining normal and abnormal beats. In future, researchers may develop classifiers by grouping heartbeats in the manner recommended by Association for the Advancement of Medical Instrumentation (Mark & Wallen 1987). In this regard, researcher may develop algorithms utilising efficient optimised training data-set and features. Authors intend to undertake this study as a future extension of the current work.