A Classification and Novel Class Detection Algorithm for Concept Drift Data Stream Based on the Cohesiveness and Separation Index of Mahalanobis Distance
Algorithm 1
Classification and novel class detection algorithm based on Mahalanobis distance.
Input:Data block , Classifier set , Nearest neighbor , Threshold
Output: Updated classifier set
(1)
for Each instance in block do
(2)
Classify (,)
(3)
if is an R-outlier for all classifiers in the classification set then
(4)
Add to the set
(5)
end if
(6)
end for
(7)
Clustering by -means () and creating a cluster point for each cluster
(8)
for Each cluster in do
(9)
Compute MN-NSC ()
(10)
if MN-NSC () is greater than 0 then
(11)
count = count + 1
(12)
end if
(13)
end for
(14)
if count greater than then
(15)
Put all instances belonging to novel class in block into class
