Final Report for CS5611 (Fuzzy Sets: Theory and Applications):

Input Selection of ANFIS, with Application to LD Prediction

ªL°ê·Ó(d857709, °Ê¾÷³Õ¤@)
ªô¥¿¶¯(g853742, °Ê¾÷ºÓ¤@)
¼ï¨Î¾Ð(g853747, °Ê¾÷ºÓ¤@)

Table of Contents

Abstract	Problem Definition	Source Information	Our Approach
Simulation Results	Conclusions	Computer Programs	Division of labor
References

Abstract

This report describes our attempt to apply ANFIS (Adaptive Neuro-Fuzzy Inference Systems) for the prediction LD (liver disorder) of people. We use two methods to reduce the input dimensions,then apply to ANFIS for prediction.

Problem Definition

This project tries to solve the problem of LD prediction using the ANFIS approach.The problem of LD prediction is concerned with the use of attributes of a specific database, such as its mcv (mean corpuscular volume), alkphos (alkline phosphotase),sgpt (alamine aminotransferase), sgot (aspartate aminotransferase),gammagt (gamma-glutamyl transpeptidase), drinks (number of half-pint equivalents of alcoholic beverages drunk per day) to estimate the healthy of liver.

Data Set Description

• Origin: This dataset was original from the BUPA Medical Research Ltd. Now that data set is also kept at UCI Machine Learning Repository. FTP access to the data set is via ftp://ftp.ics.uci.edu/pub/machine-learning-databases/liver-disorders/.
• Description:The first 5 variables are all blood tests which are thought to be sensitive to liver disorders that might arise from excessive alcohol consumption. Each line in the bupa.data file constitutes the record of a single male individual.It appears that drinks>5 is some sort of a selector on this database.See the PC/BEAGLE User's Guide for more information.
• Attribute Information:
• Input 1: mcv (mean corpuscular volume)
• Input 2: alkphos (alkline phosphotase)
• Input 3: sgpt (alamine aminotransferase)
• Input 4: sgot (aspartate aminotransferase)
• Input 5: gammagt (gamma-glutamyl transpeptidase)
• Input 6: drinks (number of half-pint equivalents of alcoholic beverages drunk per day)
• Past Usage:
• None known other than what is shown in the PC/BEAGLE User's Guide (written by Richard S.Forsyth.)

Approach

Our approach to this problem can be explained in two aspects:

• Input Selection:

Method-1:Principle Component Analysis

Method-2:ANFIS for nonlinear regression

• ANFIS Training

Simulation Results

• Data Preprocessing:
• Missing values : none
• Input normalization: none.
• Output normalization: none
• Data reduction: none
• Others: none
• Input selection:

Principle Component Analysis:(input:1,2,6)

• mcv (mean corpuscular volume)
• alkphos (alkline phosphotase)
• drinks (number of half-pint equivalents of alcoholic beverages drunk per day)

ANFIS for nonlinear regression(input:1,2,3)

• mcv (mean corpuscular volume)
• alkphos (alkline phosphotase)
• sgpt (alamine aminotransferase)
• Data utilization:
• Training data: Odd-indexed entries
• Test data: Even-indexed entries
• ANFIS Structure:
  
  • 3-input 8-rule ANFIS
• Training method:
  ANFIS applies the least-squares method and the back-propagation gradient descent for identifying linear and nonlinear parameters, respectively, in a Sugeno-type fuzzy inference systems.
• Training Results:
• Input selection plot(nonlinear regression):
  

Input selection curves.




• Membership Functions before and after learning:
  

Membership Functions before and after learning




• Average Percentage error:
  

Average Percentage error




• Eigenvalues:
  

eigenvalue




• Principal Component Analysis:(MF)
  

Principal Component Analysis:(MF)




• Principal Component Analysis:(Average Percentage error)
  

Principal Component Analysis:(Average Percentage error)




• Principal Component Analysis:(MF)other
  

Principal Component Analysis:(MF)other




• Principal Component Analysis:(Average Percentage error)other :
  

Principal Component Analysis:(Average Percentage error)other




• forget the process
  

forget the process




Concluding Remarks and Future Work

• Concluding Remarks:
  When selecting inputs in method-1(1,2,6),we found the differences compared to method-2(1,2,3) is little.
  According to RMSE,the method-1 is better
  
• Future Work:
  If we have more time and energy,we can try other training methods,such as: Genetic algorithm;Simulated Annealing...

Computer Programs

• All the computer programs for this project are available at:
• ANFIS for nonlinear regression code1
• ANFIS for nonlinear regression code2
• Principle Component Analysis
• ANFIS Training for nonlinear regression
• ANFIS Training for Principle Component Analysis

Division of Labor

• ªL°ê·Ó: Project presentation,discussion
• ªô¥¿¶¯ MATLAB Coding,discussion
• ¼ï¨Î¾Ð : Report writing,discussion

References

• [Jang97] J.-S. R. Jang, C.-T. Sun, and E. Mizutani, `` Neuro-Fuzzy and Soft Computing: A Computational Approach to Learning and Machine Intelligence,'' Prentice Hall, 1997.

Special thanks: Sir Roger Jang