Software Metrics

1
Software Metrics Dr. Edward B. Allen and Amit
Phadke allen, aap50 _at_cse.msstate.edu
3. Sources of Data
2. Benefits
1. Why Software Metrics?

• Measurement of software artifacts
• Surrogate for cognitive complexity
• Model to predict reliability

• Configuration Management System
• Bug tracking databases
• Source code

• Identify high-risk software modules
• Focus effort on in-house software
• Evaluate open-source software

4. Example Raw Metrics
5. Metric Types

• LOC Lines of Code
• V (G) McCabe Complexity
• (Decision nodes 1)
• FilIncNbr Number of included files
• RtnStmDecTypeNbr Number of data declarations

• Counting based
• Count features
• Includes redundancy
• All items are equal

• Information theory based
• Measure information
• No redundancy
• Cognitive load

6. Modeling Techniques
8. Case Studies
7. Example Model

• Past
• MPICH Predicting code churn
• PETSc Predicting modules with faults
• Potential
• Developed by collaborators
• Open Source

• Statistical Methods
• Multiple linear regression
• Poisson regression
• Logistic regression

• Machine Learning Methods
• Neural Networks
• Genetic Algorithms
• Decision trees

Logistic Regression Model log (p / (1-p)) -
0.57 0.68 FACTOR1 0.46 FACTOR2
0.39 FACTOR3 0.19 FACTOR4
0.21 FACTOR6 0.39 FACTOR7
p is probability that the file is
fault-prone Factor is a combination of raw
metrics
Empirical Software Engineering Research