Class SVM<T>

java.lang.Object

smile.model.svm.KernelMachine<T>

smile.anomaly.SVM<T>

Type Parameters:

T - the data type of model input objects.

All Implemented Interfaces:

Serializable

public class SVM<T>
extends KernelMachine<T>

One-class support vector machines for novelty detection. One-class SVM relies on identifying the smallest hypersphere consisting of all the data points. Therefore, it is sensitive to outliers. If the training data is not contaminated by outliers, the model is best suited for novelty detection.

The KernelMachine.score(Object) method (inherited from KernelMachine) returns the raw decision function value: positive values indicate inliers (normal observations) and negative values indicate anomalies. This convention is the opposite of IsolationForest, where higher scores signal anomalies.

References

1. B. Schölkopf, J. Platt, J. Shawe-Taylor, A. J. Smola, and R. C. Williamson. Estimating the support of a high-dimensional distribution. Neural Computation, 2001.
2. Jia Jiong and Zhang Hao-ran. A Fast Learning Algorithm for One-Class Support Vector Machine. ICNC 2007.

See Also:

• Serialized Form

Nested Class Summary

Nested Classes

Modifier and Type	Class	Description
static final record	SVM.Options	SVM hyperparameters.

Constructor Summary

Constructors

Constructor	Description
SVM(MercerKernel<T> kernel, T[] vectors, double[] weight, double b)	Constructor.

Method Summary

Modifier and Type	Method	Description
static <T> SVM<T>	fit(T[] x, MercerKernel<T> kernel)	Fits a one-class SVM.
static <T> SVM<T>	fit(T[] x, MercerKernel<T> kernel, SVM.Options options)	Fits a one-class SVM.
boolean	predict(T x, double threshold)	Predicts whether a sample is an anomaly.
double[]	score(T[] x)	Returns the decision function values for an array of samples.

Methods inherited from class KernelMachine

intercept, kernel, score, toString, vectors, weights

Methods inherited from class Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, wait, wait, wait

Constructor Details

SVM

public SVM(MercerKernel<T> kernel,
 T[] vectors,
 double[] weight,
 double b)

Constructor.

Parameters:

kernel - Kernel function.

vectors - The support vectors.

weight - The weights of instances.

b - The intercept;

Method Details

fit

public static <T> SVM<T> fit(T[] x,
 MercerKernel<T> kernel)

Fits a one-class SVM.

Type Parameters:

T - the data type.

Parameters:

x - training samples.

kernel - the kernel function.

Returns:

the model.

fit

public static <T> SVM<T> fit(T[] x,
 MercerKernel<T> kernel,
 SVM.Options options)

Fits a one-class SVM.

Type Parameters:

T - the data type.

Parameters:

x - training samples.

kernel - the kernel function.

options - the hyperparameters.

Returns:

the model.

score

public double[] score(T[] x)

Returns the decision function values for an array of samples. Positive values indicate inliers; negative values indicate anomalies.

Parameters:

x - the samples.

Returns:

the decision function values.

predict

public boolean predict(T x,
 double threshold)

Predicts whether a sample is an anomaly. A sample is considered an anomaly when its decision function value is below threshold. Use threshold = 0.0 for the natural SVM decision boundary.

Parameters:

x - the sample.

threshold - the decision value threshold below which the sample is declared an anomaly.

Returns:

true if the sample is predicted as an anomaly.