Predictive Models

Predictive Models

1. Defect Cost Model

   • Predicting the Flow of Defect Correction Effort using a Bayesian Network Model
     • Characteristics
       • high availability and reliability to support safety critical applications
       • real-time operation
       • code understandability and maintainability to enable multiple, distributed engineering teams
       • code portability to support the product line approach for variant handling,
       • network operation to support distributed systems
       • code efficiency due to limitations of processor and memory resources
     • Model Data
       • Change entry describes the development of a newly added, modified or removed requirement. It holds information about the development effort needed for this change.
       • Defect Analysis entry describes a defect in an existing feature. It stores the information about the analyzing effort, where the defect has been detected (origin phase) and in what phase it has been corrected (correction phase).
       • Defect Correction describes what changed to solve the defect. It stores information about the correction effort
     • Defect Cost Model
       • How much development effort is spent?
       • How error-prone is the development?
       • How much effort is spent on QA measures?
       • How effective are these QA measures?
       • What is the effort involved for defects shifting from one phase to another?

2. Support Vector Regression

   • How Effective is Tabu Search to Configure Support Vector Regression for Effort Estimation?
     • Tabu Search
       • defining a representation of possible solutions and the way to generate the initial one
       • defining local transformations (i.e., moves) to apply to the current solution in order to explore the neighbor solutions
       • choosing a means to evaluate the neighborhood (i.e., an objective function);
       • defining the Tabu list size, the aspiration criteria, and the termination criteria
     • Support Vector Regression
       • Large differences in the ranges of the features’ values can have the unwanted effect of giving greater importance to some characteristics than to others. To address this issue, a data preprocessing step should be applied.
       • Usually data in the input space are non-linear, and thus kernel functions should be considered to map a problem in a feature space where the target function consists of a line
     • Variables
       • Number of programming languages adopted in the project
       • Number of Developers involved in the project
       • Mean number of years of experience for the team members.
       • Effort in person-hours
       • Total number of functionality (new and reused)
       • Total number of new functionaity
       • Number of features/functions reused without any adaptation
       • Number of reused high-effort features/ functions adapted
       • Number of new high-effort features/ functions
       • Total number of high-effort features/ functions.
       • Number of reused low-effort features adapted
       • Number of new low-effort features/functions.
       • Total number of low-effort features/ functions

3. Software defect prediction

   • The limited impact of individual developer data on software defect prediction
     • Prediction Model and Previous Results
       • Negative Binomial Model
       • Evaluation Metrics
       • Previous Results
     • Does Data on Developer Counts Improve Prediction Accuracy?
       • vPrediction Results without Developer Information
       • Impact of Developer Metrics