Support-Vector Machine

We want to train a kernel SVM for banknote authentication. Without kernel the decision surface will be a hyperplane more or less identical to the hyperplane obtained from logistic regression. Separation by a hyperplane works, but for both classes there are samples very close to the hyperplane. Looking for nonlinear separation should yield fewer ambiguous samples.

Load Data Set

Task: Load the banknotes data set (cf. Detecting Forgery with k-NN). Drop the entropy column.

# your solution

SVM

Task: Use Scikit-Learns’s SVC to create a model for banknote authentication. Try different kernels (RBF, polynomials of different degrees). Use hyperparameter optimization for choosing the parameter C.

Solution:

# your solution

Decision Surface

Task: Train a second model based on variance and skewness only. Plot the decision surface (the surface separating the classes).Highlight all support vectors.

Solution:

# your solution