Comparison of various models and optimum range of its parameters used in SVM classification of digital satellite image

SVM is a one of the most favorite and efficient supervised-classification technique. As it is supervised so learning of the model directly affects the result. Learning of SVM is done with the help of sample data points and model; it can be either linear or non-linear. Number of models has been used so far for multiclass classification but which one is better with optimal value of parameters? If we have several models for SVM and each model has some parameters so one could confuse about to select the model and optimal values of its parameters. SVM is broadly used for binary classification or multiclass classification with linear model. In this article we will describe linear as well as non-linear models and their accuracy with focus on the optimal value of parameters used in each model and tolerating cost of SVM. In this paper class agreement between model and particular class is explained so that one can understand better that which model is responding well for which particular class. So this paper will provide comparative study between various models, for hyper-plane, used with SVM for LULC classification of digital satellite image. The kappa coefficient is calculated to compare the classified result and standard error is evaluated under the 95% confidence interval to estimate the error in result.