SOFTWARE METRICS

1
SOFTWARE METRICS

• Software Metrics Roadmap
• Norman E Fenton and Martin Neil
• Presented by
• Santhosh Kumar Grandai

2
OVERVIEW

• What is Software Metric?
• About Software Metrics
• Regression Models
• Causal Models - Bayesian Belief Net(BBN)
• Why do we need Causal Model?
• Conclusion

3
SOFTWARE METRIC

• Metrics is Measurement.
• Various Metrics on various Phases of life cycle
  model.
• Purpose To detect problems early in the
  software process.

4
About Software Metrics..

• Over 30 years old.
• Mid 1960s LOC used as the basis for measuring
  Programming productivity and Effort.
• Has Two Components.
• 1. Component Concerned with defining the
  actual measures.
• 2. Component Concerned with how we
  collect,manage
• and use the measures.

5
About Software Metrics
6
About Software Metrics.

• External Attributes
• - Ones interested to know about.
• Internal Attributes
• - Control and Measure Directly.
• To Predict Effort/Cost of development Process.
• To Predict quality of software products.

7
About Software Metrics and Regression Models

• First Key Metric was Lines of code(LOC).
• Quality Prediction
• - Defect Density
• Quality and Effort

Quality
Product Size
Effort/Cost
8
REGRESSION MODELS

• Does not support quantitative managerial decision
  making during software life cycle
• - No support for risk assessment
• - No support for risk reduction
• Misunderstanding between cause and effect.
• Does not consider causality,uncertainty,evidence.

9
REGRESSION MODEL

• An practical Example,
• - Data on car accidents show that Jan and Feb
  are the months will fewest fatalities.
• - An regression model is built from available
  data.
• - No causal relationship.
• - Sensible decision about safest time to drive
  cannot be made.

Number Of Fatalities
Month
10
CAUSAL MODEL

• An practical Example,

11
CAUSAL MODEL

• In Software Metrics
• - Dominated by Regression Models.
• - Need causal Models.

Problem complexity
size
Resource quality
schedule
Effort
Product Quality
12
CAUSAL MODEL

• Can give answers to questions
• - For a specification of this complexity,and
  given these limited resources,how likely I am
  achieve a product of suitable quality?
• - How much can I scale down the resources if I
  am prepared to put up with a product of specified
  quality?.
• Regression models cannot.

13
Analysis

• We see that only defect counts are being used in
  measure of quality. Not true
• Consider Hypothesis
• Suppose you know that a large number of
  defects are found in a software module prior to
  release.Is it likely that this module will reveal
  many defects post-release?.
• - Yes, Popularly believed.
• - Empirical Evidence shows it is an invalid
  hypothesis.

14
Analysis
15
Analysis

• Modules with high pre release faults had less
  post-release faults,
• - The amount of testing must be incorporated
  into any
  predictive module of defects.
• - Operational usage must also be
  incorporated.

16
REGRESSION MODELS

• Regression Models,
• - cannot consider resourcing constraints.
• - cannot handle uncertainty.
• - no cause and effect relationship.
• Not suitable for risk assessment and reduction.

17
CAUSAL MODEL

• Causal models can handle,
• - Diverse process and product variables.
• - Genuine cause and effect relationship.
• - Empirical evidence and expert
  judgement.
• - Uncertainty.
• It covers the crucial concepts missing from the
  classical regression-based approaches.

18
BBN

• Bayesian Belief Nets(BBN) is a type of causal
  model,which uses Bayesian probability.
• BBN is a graphical network together with the
  associated set of probability tables.
• - Nodes represent Uncertain values.
• - Causal relationship.

19
BBN

• To predict defect counts for software modules.

20
BBN
21
BBN

• For given Input pre-release defects(less than 10)
  and many post-release(between 30 and 40),
• Output is very low amount of testing was
  done.
• Given the Evidence of a variable BBN calculates
  the Probability of each state.

22
BBN

• Absence of BBN for a long time
• - No proper algorithm.
• - No software tool.
• Hugin tool is used.

23
BBN

• Used to,
• - Provide safety or reliability arguments
  for critical systems.
• - Provide improved reliability
  predictions of prototype
• military vehicles.
• - Provide predictions of insurance risk
  and operational risk.
• Drawback,
• - cannot be used in decision making in
  deployment of systems .
• -lacks political,financial,environmenta
  l criteria.
• - Multi Criteria Decision Aid(MCDA) deals
  with the above criteria.
• - Deployment of system combination of
  two.

24
BBN

• Technology Transfer
• - Project managers are more likely to use this
  model for decision making.
• - They do not understand the underlying
  theory.
• - Provide simple,configurable front ends.

25
Conclusion

• Statistical models do not provide decision
  support for risk assessment and reduction.
• Causal models like BBN do provide decision
  support for risk assessment and reduction.
• Organizations that collect basic metrics data and
  follow defined life-cycles,will be able to apply
  causal models effectively.

26
References

• Software Metrics Roadmap,Norman E Fenton
  Martin Neil,Computer Science Department,Queen
  Mary and Westfield College,London.
• http//www.dcs.qmul.ac.uk/norman/BBNs/BBNs.htm

27

• Thank you!!!
• Questions ?