A hybrid SVM based decision tree

Preview:

DESCRIPTION

A hybrid SVM based decision tree . Presenter: Tsai Tzung Ruei Authors: M. ArunKumar n, M.Gopal. 國立雲林科技大學 National Yunlin University of Science and Technology. PR.2010. Outline. Motivation Objective Methodology Experiments Conclusion Comments. Motivation. - PowerPoint PPT Presentation

Citation preview

Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.

A hybrid SVM based decision tree

Presenter: Tsai Tzung Ruei Authors: M. ArunKumar n, M.Gopal

PR.2010

國立雲林科技大學National Yunlin University of Science and Technology

Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.Outline

Motivation Objective Methodology Experiments Conclusion Comments

2

Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.Motivation

SVMs are considerably slower in testing phase than other techniques. This is because the computational complexity of SVM’s decision function scales with respect to the number of support vectors. Hence if the number of support vectors is very large, SVMs will take more time to classify a new datapoint.

3

Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.Objective

To proposed a hybrid SVM based decision tree to speedup SVMs in its testing phase for binary classification tasks.

4

The objective To predict whether a household has an income greater than $50 k.

The outcomes(1) DTs are much faster than SVMs in classifying new instances.(2) SVMs perform better then DTs in terms of classification accuracy.

Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.Methodology

5

SVMDT

SVMDT

Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.Methodology

SVMDT algorithm

6

Train_data: set of training datapoints Train_target: corresponding target for Train_data New_target: targets to be used for DT training

Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.Methodology

SVMDT algorithm

7

Class3

Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.Methodology

8

Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.Experiments

Adult datasets

9

Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.Experiments

Checkerboard dataset

10

The result:The classification accuracy of SVMDT was same as that of SVM

Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.Experiments

SVMDT results on other binary datasets

11

Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.Experiments

SVMDT comparison with FVS

12

Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.Conclusion

MAJOR CINTRIBUTION On all the datasets, SVMDT has shown impressive results with

significant speedup when compared to SVM, without any compromise in classification accuracy.

FUTURE WORK To realize the potential of SVMDT in multiclass classification.

13

Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.Comments

Advantage Create a novel way of decreasing testing time of SVMs and it does not

contradict with the existing approaches.

Drawback ……

Application classification

14