This project involves the implementation of efficient and effective RBF SVC on MNIST data set. The MNIST data comprises of digital images of several digits ranging from 0 to 9. Each image is 28 x 28 pixels. Thus, the data set has 10 levels of classes.
RBF SVC seems to work better on a scaled dataset, hence, the train and test sets were scaled between zero and one by dividing through by 255 (which is the maximum value in the dataset)
Gamma is a parameter of the radial basis function kernel, a small gamma in RBF means high standard deviation around each point and vice-versa.
On the other hand, the cost is a parameter of the SVM. Cost determines how much penalty is given to data point on the wrong side of the hyperplane. A Low cost simply implies a smooth decision surface. inversely, a high cost aims at correctly classifying all the training set, which might lead to a complex model.