This subsection describes the evaluation measures used
in this experiment. The evaluation measures are the following:
1) Features Predictive Performance
In order to measure predictive performance of extracted
features we uses the information gain with respect to the
class.
The Information Gain Attribute Evaluation (InfoGain Attribute
Evaluation) is a method that evaluates the worth of an
attribute by measuring the information gain with respect to
the class [25]. This method is able to evaluate the predictive
power of an attribute (an extracted feature in our case). Accordingly,
we use the method to identify the influence of a
feature in UML CD prediction. The InfoGain Attribute Evaluation
produce a value from 0 to 1 in which a higher value
indicates a stronger influence.
2) Classification Algorithm Performance
We use a confusion matrix to evaluate the machine learning
classification algorithms. Table II shows a confusion
matrix. In this table, for the case of the actual data is positive
(Y), TP represents the number of correct predictions (true
positive) and FN represents the number of incorrect predictions
(false negative) by the classification algorithms. In the
case of the actual data is negative (N), FP represents the
incorrect predictions (false positive) while TN represents
correct predictions (true negative).