Feature selection for SPC chart pattern recognition using fractional factorial experimental design

Feature selection is one of the important steps in designing a pattern recognizer. This paper presents a study to select a minimal set of statistical features for SPC chart pattern recognition using the fractional factorial experimental design. A resolution IV design was used to identify the significant features from a list of ten possible candidates to represent the input data streams. Further judgment was adopted to arrive at the final selection in the light of some ambiguities among confounded two-factor interactions. The final six selected features set comprising, autocorrelation, cusum, mean, standard deviation, mean-square value, and skewness as the input vector resulted in an average correct classification rate of 97.1% and standard deviation of 0.878. The methodology adopted in this study could be applied to other feature selection problems beside for SPC chart pattern recognition.

NOTE: TO READ THIS AND OTHER IPROMS 2006 PAPERS, PLEASE REGISTER FOR THE CONFERENCE.

REGISTRATION IS FREE.

CLICK here TO REGISTER.

Attachment	Size
PID208983.pdf	189.66 KB

Submitted by Ebubekir on Mon, 26/06/2006 - 4:33pm.
login or register to post comments |

Hello Dr Adnan

Submitted by Pham on Wed, 05/07/2006 - 7:40pm.

Hello Dr Adnan,

Good to read your paper.

I was interested to see some of the features you used.

I would have placed more weight on those features with a geometric significance.

I believe they are the features humans employ when recognising patterns of the kind we try teaching machines to identify.

Best wishes.

DTP.

Thank You Prof. Pham

Submitted by adnan on Thu, 06/07/2006 - 9:26am.

Hello Professor Pham,

Thank you for your interest in my paper.

Actually, we have published the mathematical expressions for these statistical features. Please refer to paper ref no. 4 (Hassan et al, 2003).

As you may notice, this work was motivated by your earlier work (Pham and Wani, 1997)(ref no.8) which used nine geometric features. We decided to us statistical features since they are established measures for variability. In fact when we plotted the data streams interm of bar graphs, we can observe graphical shapes produced by each features. We have observed that from these graphs, different features possess different discrimination power. Thus, the statistical features that we used in this paper can also be transformed into graphical shapes (graphs).

I hope this explaination answer the above.

Thank you.
Adnan.

Thank you, Dr Adnan!

Submitted by Pham on Thu, 06/07/2006 - 4:42pm.

Thank you, Dr Adnan for the very clear explanation. Our best wishes to you and all our friends in Malaysia! We look forward to collaborating with you again.

See you soon.

DTP.

Thanks!

Submitted by adnan on Fri, 07/07/2006 - 3:59am.

Dear Professor,
Thank you. We are fine here. UTM new semester/classes will start next week monday 10 July 2006.

We have benefitted from earlier collaborations, APOST (2002-2003) and EAPSTRA project (2004-2005). Thanks also to the European Commission for supporting and sponsoring the projects.

We also look forward to collaborate with you again.

Cheers.
Adnan.

Collaboration

Submitted by Pham on Fri, 07/07/2006 - 9:48pm.

Dear Dr Adnan,
Thank you. You were a very good project partner. When there are new opportunities, we shall let you know.
Best wishes.
DTP.

Feature selection for SPC chart pattern recognition using fractional factorial experimental design

Comment viewing options

User login