Virtual Labs

Predicting Credit Card Fraud using Support Vector Machine

What is the main advantage of the kernel trick in Support Vector Machines?

a: It explicitly transforms the data into a very high-dimensional space Explanation

Explanation

b: It allows non-linear decision boundaries without ever computing coordinates in the high-dimensional space Explanation

Explanation

c: It reduces the number of features in the dataset Explanation

Explanation

d: It automatically removes outliers Explanation

Explanation

In highly imbalanced credit card fraud datasets, which of the following is generally the most informative summary metric?

a: Overall accuracy Explanation

Explanation

b: ROC-AUC Explanation

Explanation

c: Precision-Recall AUC (PR-AUC) Explanation

Explanation

d: Mean Squared Error Explanation

Explanation

What does a very large value of the regularization parameter C indicate in soft-margin SVM?

a: The model prioritizes a wide margin even if many points are misclassified Explanation

Explanation

b: The model tries to classify nearly all training points correctly, accepting a narrower margin Explanation

Explanation

c: The model ignores the margin and only minimizes hinge loss Explanation

Explanation

d: The model switches to a hard-margin formulation Explanation

Explanation

Which of the following statements about support vectors is true?

a: They are all the points inside the margin Explanation

Explanation

b: They are the points that lie exactly on or within the margin boundaries and determine the hyperplane Explanation

Explanation

c: They are randomly selected training examples Explanation

Explanation

d: They are only the correctly classified points Explanation

Explanation

Why is the RBF (Gaussian) kernel typically the best-performing choice for credit card fraud detection with SVM?

a: It is the fastest to compute Explanation

Explanation

b: It assumes linear separability Explanation

Explanation

c: It can create highly flexible, localized decision boundaries suitable for complex fraud patterns Explanation

Explanation

d: It requires no hyperparameter tuning Explanation

Explanation

Which preprocessing step is considered absolutely essential before training an SVM (especially with RBF kernel) on fraud data?

a: Feature scaling / standardization Explanation

Explanation

b: Converting all features to categorical Explanation

Explanation

c: Removing 99% of legitimate transactions Explanation

Explanation

d: Applying PCA to reduce to 2 dimensions Explanation

Explanation