Support Vector Data Description (SVDD)

Python Code for abnormal detection or fault detection using SVDD.

Email: iqiukp@outlook.com

Main features

SVDD model for training dataset containing only positive training data. (SVDD)
SVDD model for training dataset containing both positive training data and negative training data. (nSVDD)
Multiple kinds of kernel functions.
Visualization module including ROC curve plotting, test result plotting, and decision boundary.

Requirements

matplotlib
cvxopt
scipy
numpy
scikit_learn

About SVDD model

Two types of SVDD models are built according to the following references:

[1] Tax D M J, Duin R P W. Support vector data description[J]. Machine learning, 2004, 54(1): 45-66.

A simple application for decision boundary (using differnent kernel functions)


# -*- coding: utf-8 -*-

import sys
sys.path.append("..")
from src.svdd import SVDD
from src.visualize import Visualization as draw
from data import PrepareData as load

# load banana-shape data
trainData, testData, trainLabel, testLabel = load.banana()


# kernel list
kernelList = {"1": {"type": 'gauss', "width": 1/24},
              "2": {"type": 'linear', "offset": 0},
              "3": {"type": 'ploy', "degree": 2, "offset": 0},
              "4": {"type": 'tanh', "gamma": 1e-4, "offset": 0},
              "5": {"type": 'lapl', "width": 1/12}
              }


for i in range(len(kernelList)):

    # set SVDD parameters
    parameters = {"positive penalty": 0.9,
                  "negative penalty": 0.8,
                  "kernel": kernelList.get(str(i+1)),
                  "option": {"display": 'on'}}
    
    # construct an SVDD model
    svdd = SVDD(parameters)
    
    # train SVDD model
    svdd.train(trainData, trainLabel)
      
    # test SVDD model
    distance, accuracy = svdd.test(testData, testLabel)
    
    # visualize the results
    # draw.testResult(svdd, distance)
    # draw.testROC(testLabel, distance)
    draw.boundary(svdd, trainData, trainLabel)

gaussian kernel function

linear kernel function

polynomial kernel function

sigmoid kernel function

laplacian kernel function

A simple application for abnormal detection or fault detection


# -*- coding: utf-8 -*-

import sys
sys.path.append("..")
from src.svdd import SVDD
from src.visualize import Visualization as draw
from data import PrepareData as load

# load banana-shape data
trainData, testData, trainLabel, testLabel = load.iris()

# set SVDD parameters
parameters = {"positive penalty": 0.9,
              "negative penalty": 0.8,
              "kernel": {"type": 'gauss', "width": 1/24},
              "option": {"display": 'on'}}


# construct an SVDD model
svdd = SVDD(parameters)

# train SVDD model
svdd.train(trainData, trainLabel)


# test SVDD model
distance, accuracy = svdd.test(testData, testLabel)

# visualize the results
draw.testResult(svdd, distance)
draw.testROC(testLabel, distance)

test result