Calibrating Function Points Using NeuroFuzzy Technique

1
Calibrating Function Points Using Neuro-Fuzzy
Technique
Luiz F Capretz
Danny Ho
Vivian Xia
IT Department HSBC Bank Vancouver, BC
Canada Vivian_xia_at_hsbc.ca
Department of Electrical and Computer
Engineering University of Western
Ontario London, Ontario, Canada lcapretz_at_eng.uwo.c
a
NFA Estimation Inc. London, Ontario,
Canada danny_at_nfa-estimation.com
2
Roadmap

• Concepts of Calibration
• Neuro-Fuzzy Function Points Calibration Model
• Validation Result
• Conclusions

3
Calibration Concept

• Internal Logical File (ILF) Complexity Matrix

DET, RET --- Component Associated Files

• Same methodology for all FP 5 components

External Input, External Output, External
Inquiry Internal Logical File, External
Interface File
4
Calibration Concept Contd

• e.g. One project has 3 Internal Logical Files
  (ILF)

• Calibrate complexity degree by fully utilizing
  the number of component associated files
• Calibrate to fit specific application

5
Calibration Concept Contd

• Unadjusted Function Points Weight Values
• UFP weight values are determined in 1979 based on
  
• Albrechts study of 22 IBM Data Processing
  projects

• .
• Calibrate UFP weight values to reflect global
  software industry trend

6
Neural Networks Basics
Learning from Data Source

• Adapting capability
• Modeling any complex nonlinear relationships
• Lack of explanation black box
• Cannot take linguistic information directly

7
Neuro-Fuzzy Function Points Calibration Model
Overview
Estimation Equation
ISBSG 8
Project Data
Validated for better estimation
Calibrated by Neural Network
MMRE, PRED
Calibrated by Fuzzy Logic
8
Calibrating by Fuzzy Logic
9
Calibrating by Neural Network

• Learn UFP weight values by effort
• the values should reflect complexity
• complexity proportioned to effort
• 15 UFP inputs as neurons
• Back-propagation algorithm

10
Data Source --- ISBSG Release 8

• ISBSG
• International Software Benchmarking Standards
  Group
• Non-profit organization
• Release 8 Contains 2,027 projects
• 75 built in recent 5 years
• Filter on ISBSG 8 data set
• Filter Criteria
• Quality, Counting method, Resource level,
• Development Types, UFP breakdowns
• Shrink to 184 projects

11
Validation Methodology

• Developed a calibration tool
• Randomly split data set
• totally 184 data points
• 100 training points
• 84 testing points for validation
• Repeat 5 times
• Using estimation equation for comparison

12
Validation Results (MMRE)

• MMRE
• Mean Magnitude of Relative Error
• Criteria to assess estimation error
• The lower the better

13
Validation Results (PRED)

• PRED
• Prediction at level p
• Criteria to assess estimation ability
• The higher the better

14
Conclusions

• Neuro-Fuzzy Function Points model improves
  software cost estimation by an average of 22.
• Fuzzy logic calibration part improves UFP
  complexity classification.
• Neural network calibration part overcomes
  problems with UFP weight values.