AI-ordlista
Den kompletta ordlistan över AI
Separating hyperplane
An N-1 dimensional subspace in an N-dimensional space that maximizes the margin between normal data and the origin, serving as a classification boundary.
Support vectors
Critical points that define the decision boundary and support the hyperplane, being the closest to the boundary in the transformed space.
RBF Kernel
Gaussian radial basis function kernel that projects data into an infinite space, enabling the detection of complex non-linear anomalies.
Nu-SVM
Variant of One-Class SVM using the nu parameter to control the expected ratio of anomalies, offering more intuitive control over the error rate.
SVDD
Support Vector Data Description, an alternative method that seeks a minimal hypersphere enclosing normal data rather than a separating hyperplane.
Contamination rate
Parameter estimating the proportion of anomalies in the dataset, influencing the position of the decision boundary in the algorithm.
Feature space
Multidimensional space where data is represented after transformation by the kernel function, enabling better class separation.
Semi-supervised learning
Hybrid approach where One-Class SVM is trained on labeled normal data to detect unlabeled anomalies in production.
Novelty detection
Specific application of One-Class SVM to identify new classes or behaviors never observed during training.
Confidence quantile
Statistical threshold based on the distribution of anomaly scores, allowing to calibrate the confidence level in detection.
One-class cross-validation
Validation technique specific to anomaly detection algorithms preserving the one-sided nature of data during evaluation.