SSC voting rule

Syntax

       [ID, FVP] = SSC(dp)

[ID, FVP] = SSC(dp, filterflag)

[ID, FVP] = SSC(dp, filterflag, alpha)

 Description

[ID, FVP] = SSC(dp) returns the classified labels ID and the final voting probability (FVP) vectors of all potential class labels by the SSC algorithm.

       Decision profile, dp, is a 3-dimension matrix organized as shown in Fig. 1. A modified logistic filter is used and alpha is 0.1 by default. (see equation (29) in reference [1])

                                                                          

                                                                                      Fig. 1 The organization of dp.

[ID, FVP] = SSC(dp, filterflag) return the classified labels ID and the final voting probability vectors of all potential class labels by the SSC algorithm.

If filterflag is true or 1, a modified logistic filter is used and the default value of alpha is 0.1. (see equation (29) in reference [1])

If filterflag is false or 0, a modified logistic filter is not used. (see equation (28) in reference [1])

 

[ID, FVP] = SSC(dp, filterflag, alpha) return the classified labels ID and the final voting probability vectors of all potential class labels by the SSC algorithm.

If filterflag is ture or 1, a modified logistic filter is used and the value of alpha is alpha as the user specified (alpha is a positive real number) (see equation (29) in reference [1])

If filterflag is false or 0, a modified logistic filter is not used. (see equation (28) in reference [1])

 

Example

Example 1 (Single Instance)

A 3-class classification problem with 4 hypotheses as shown in Fig. 2. (This example can also be found in reference [1]. See Section III-B.)

                                                                 

                                                                                    Fig. 2 Decision profile of example 1

[id, fvp] = SSC(dp)

id =

     1

fvp =

     0.7671   0.0404    0.1359

In this example, the testing instance is voted as class 1 label. The final voting probabilities for all potential classes/labels are :

Example 2 (Multiple Instances)

A 4-class 5-instance classification problem with 3 classifiers/hypotheses as shown in Fig. 3.

                                      

                                                                                              Fig. 3 Decision profile of example 2

       

[id, fvp] = SSC(dp)

 id =

     4

     3

     3

     1

     2

fvp=

    0.2055    0.0609    0.0422    0.2542

    0.2641    0.0968    0.2878    0.1399

    0.1453    0.1082    0.3146    0.0637

    0.2717    0.0798    0.1458    0.0802

    0.1456    0.2725    0.0730    0.0691

 Source code

(1) You can download the source code here.

(2) You can download the demo for example 1 here.

(3) You can download the demo for example 2 here.

Reference

The software package and examples provided here are associated with our following paper published in the IEEE TNNLS. If you are considering to use this algorithm in your research/work, please cite and refer to our following paper:

H. He and Y. Cao, "SSC: A Classifier Combination Method Based on Signal Strength," IEEE Trans. Neural Networks and Learning Systems, vol. 23, issue 7, pp. 1100-1117, 2012