Predicting Faulty Classes Using Design Metrics with Discriminant Analysis

Nowadays risk assessment is one of software engineering processes that plays important role in software development life cycle. Applying risk assessment to software the earlier is the better. Developers should detect defects of software early at design phase so the improvement action such as refactoring can be taken. Constructing fault prediction model using design metrics is one approach that can help developers to identify the faulty classes at early phase. This paper collects object-oriented design metrics and introduces some new metrics that tend to affect the existing of faults in classes, then construct the fault prediction model with discriminant analysis technique. The prediction model was trained by data collected from sale system and was validated using data from CD-selection system. The result indicates that 12 of 14 design metrics are associated with fault-proneness and the model can be used to classify faulty level of new classes.