Support Vector Machine Implementation to Separate Linear and Non-Linear Dataset

Abstract

This paper presents the implementation of separating the linear and non-linear data using support vector machine (SVM) algorithm. First let us understand what linear and non-linear datasets are. Linear datasets are the data that can be easily separable using a straight line. Such data are usually easy to implement in Artificial Neural Networks as they require a smaller number of hidden layers for its computation. Less layers implies a smaller number of weights assigned to the nodes present between the layers and less amount of time needed is needed to compute and update the weights for the current neural network. Hence linear datasets are easy to train and model. Whereas non-linear datasets are those datasets which cannot be separated by a straight line. For such datasets more hidden layers and weights are required and also more time and computational power is needed for a system to update the weights and train the model to give a better and sophisticated output data. As a result, training and modeling such neural network is tedious due to its complexity. To solve this problem SVM comes into picture. SVM stands for Support Vector Machine which is a machine or an algorithm that helps in classification and diversifying the data given to it. The data provided to the Support Vector Machine (SVM) should be a labeled one. Then these datasets are given to a training model where the training process of the neural network is being undergone. Once the training is completed, the next step is to predict the output. For this process we have to provide a new data that may or may not belong to the dataset, so that the neural network can predict the output of it. If the prediction is wrong, again the training is done until we get the actual output matching with the desired output given by the designer for verification purposes. This is the basic working process under the SVM algorithm. The linear data that is used for this separation is an Iris dataset that contains various information about the different plant-life growing from 2002-2004.