Open AccessDecision Boundary Extraction of Classifiers
Author(s) -
Ningyuan Gong,
Zixiong Wang,
Shuangmin Chen,
Liu Guo-zhu,
Shiqing Xin
Publication year - 2020
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1651/1/012031
Subject(s) - decision boundary , classifier (uml) , artificial intelligence , computer science , decision tree , support vector machine , pattern recognition (psychology) , machine learning , quadratic classifier , random subspace method , voronoi diagram , visualization , data mining , mathematics , geometry
In the field of machine learning, explicitly extracting the decision boundary of a classifier not only helps to visualize the differences between different classifiers, but also provides a more convenient way to determine the class of a sample. We proposed a boundary extraction method that relies on the output result of the classifier but not on the classifier mechanism. We find the initial decision boundary based on Voronoi diagram, and then take the smoothness and simplicity as the driving goal, adaptively adjust the shape of the boundary until convergence. In order to verify the effectiveness and usefulness of the algorithm, the decision boundaries generated by four different classifiers, ANN, SVM, Random Forest, and ELM, were visualized on the four data sets, the classification accuracy is also analyzed based on the extracted decision boundaries. The test results show that the visualization results of decision boundaries and the classification accuracy based on explicit decision boundaries are highly consistent with the classification accuracy of the classifier, which can simulate the decision mechanism of the classifier well.