8/10/2019 Vector Machine

1/2

VECTOR MACHINE

A Support Vector Machine is a binary classifier; it aims to classify two classes of

instances by finding the maximum separating hyper plane between the two (Smitha et al, 2011). For

this reason SVM tends to generalize better. With the basic design of Support Vector Machines, it

can only discriminate between two classes. In order to allow for the classification of more than twoclasses, one of the following methods can be employed. One such method is the "one-vsone"

method which creates one binary classifier for each pair of classes. If for our case, for example threeclasses are there then three binary classifiers are created. Support Vector Machines in their simpleform, are called linear classifiers. It is possible however to create a nonlinear SVM by increasing the

dimensionality of the feature space, and by using the so-called "kernel-trick". It is thus possible to

find a separating the hyper plane in a higher dimensions where such a hyper plane would not exist

in lower dimensions. There are many choices for which kernel, to use. The standard choices are the

linear kernel (which is otherwise called as dot-product kernel), the polynomial kernel and the

Gaussian Kernel. The Gaussian Kernel is the special case of RBF kernel. In the standard case, the

distance used, is the Euclidean distance. In the RBF kernel, the parameters determine, the width of

the kernel, and d(x, y) is the distance metric.

Another intelligent classification technique to identify normal and abnormal slices of

brain MRI data based on Least Squares Support Vector Machines (LS-SVM) was proposed

(Selvaraj et al., 2007). LS-SVM had a higher accuracy of classification over other classifiers. The

number of false negative in LS-SVM was very low compared to SVM and a high degree of

sensitivity of the classifier to abnormal images. Due to automatic defects detection in MR images of

brain, extensive research is being performed. A Novel automatic brain tumor detection method

using Gabor wavelets was proposed (AmirEhsan Lashkari, 2010). The neural network had beentrained using back propagation algorithm and training process was continued until the Mean Square

Error (MSE) became constant with about accuracy of 98.15%. This work has some limitations

because of using all 3 modalities T1, T2_weighted and PD MR Images. The designed brain cancerdetection and classification system by Joshi et al., (2010) use conceptually simple classification

method using the Neuro Fuzzy logic. Texture features are used in the Training of the Artificial

Neural Network. Co- occurrence matrices at different directions are calculated and Grey Level Co-

occurrence Matrix (GLCM) features are extracted from the matrices. This system providesprecision detection and classification of astrocytoma type of cancer.

In work related to content based image retrieval (CBIR), automatic x-ray image

classification was proposed with multilevel feature extraction (global, local and pixel features) used

SVM classifier. The result of accuracy with SVM was 89% when compared with KNearest

neighbour with 82% (Mueen et al., 2007).

This work compares the performance of conventional ANFIS network and extreme-

ANFIS on regression problems proposed by Jagtap and Pillai in 2014. ANFIS networks incorporate

the explicit knowledge of the fuzzy systems and learning capabilities of neural networks. This

learning technique overcomes the slow learning speed of the conventional learning techniques like

neural networks and SVM without sacrificing the generalization capability. The structure of

extreme-ANFIS network is similar to the conventional ANFIS which combines the fuzzy logic's

qualitative approach and neural network's adaptive capability. As in the case of ELM, the first layer

parameters of the proposed learning machine are not tuned. Performance on two regression

problems shows that extreme-ANFIS provides better generalization capability and faster learningspeed.

8/10/2019 Vector Machine

2/2

Neurofuzzy networks have become a powerful alternative strategy to develop fuzzy

systems, since they are capable of learning and providing IF-THEN fuzzy rules in linguistic orexplicit form. Amongst such models, ANFIS is recognized as a reference framework, mainly for its

flexible and adaptive character. In this paper, we extend ANFIS theory by experimenting with a

multi-net approach wherein two or more differently structured ANFIS instances are coupled to play

together. Ensembles of ANFIS (E-ANFIS) enhance ANFIS performance skills and alleviate some ofits computational bottlenecks. Moreover, they promote the automatic configuration of different

ANFIS units and the a posteriori selective combination of their outputs.