Support Vector Machines and Applications

What is Pattern Recognition

Pattern recognition deals with discovering the classes in a dataset. The dataset consists of objects of different classes. Objects can be images, signals or any other details of the data depending on the application. Pattern recognition can be used in various areas such as character recognition, anomaly detection, face recognition, signal classification, medical decision making etc. Pattern recognition is the fundamental building block of intelligent system.

Features

Features represent the distinguishing information retrieved from the data. This information helps to differentiate between the data from various classes. For character data the features can be pixel intensities. For signal classification the features can be energy levels at different frequency. Feature detection itself is a vast area. Features impact the classification performance. Classification model is trained learns boundaries between various classes using these features

Classification

Classification is the task of distinguishing the data into different classes to which it belongs. For classification data is preprocessed then appropriate features are extracted and depending upon the application and the data available supervised, unsupervised or semi-supervised classification is used.

Supervised, Unsupervised, and Semi-Supervised Classification

In case of supervised classification training data has the correct label for the class to which is belongs to .Features are extracted from the training data, these features along with the labels are presented to the classifier, using a learning algorithm the classifier learns the decision boundaries between various classes. Features are extracted from the test data and then presented to the classifier and the classifier predicts the label for it. Unsupervised learning is used for clustering, in case of unsupervised learning dataset no information regarding which class the data belongs to and even the number of classes dataset contains. Unsupervised learning assigns the patterns to the various classes based on the similarity of the features. Semi-supervised learning on the other hand has data points with labels as well as data points for which the class is not known. Semi-supervised helps to cluster the unlabeled data by using the constraints derived from the labeled data.

Support Vector Machine Introduction

Support vector machine is one the most powerful machine learning algorithms used for classification or regression tasks. Support Vector Machines are used in various applications such as face recognition, medical decision making, regression tasks etc. with high accuracy.

Perceptron Algorithm

If we consider a 2 dimensional data the linear discriminant function is given as

Perceptron algorithm tries to find this hyperplane and uses the hyperplane for classifying the linearly separable data. All the data points satisfying the relation lie to one side of the hyperplane, whereas data points satisfying the relation belong to other side of the hyperplane.