Nt1310 Unit 6 Case Study

that What make us identify the clusters from noise points is the high contrast in density within and outside the cluster ,where the density within a cluster is more higher than the density outside it. In order to perform DBSCAN we must understand the following concepts:

Defenition 2.3:(Eps-neighborhood) , Eps neighborhood of a point p in data is {q ∈D | dist(q , p) ≤ Eps} (Ren , Liu & Liu 2012).

Defenition 2.4:(Core point) is a point that has at least number of neighbors greater than Minpts within a specified Eps (Ren , Liu & Liu 2012).

Defenition 2.5:(Directly density reachable) any point p in D is directly density reachable from q if p is one of the neighbors of q within Eps and q is core point (Ren , Liu & Liu 2012).

Defenition

Mark any point p ∈ SD as unclassified

3. set initial value of cluster identifier cid to zero

4. cid=0

5. For i=1 to number of points

6. Foreach point p in points

7. if p.cid=unclassified then

8. if Expand cluster(points ,p , cid, Eps, Minpts) then

9. cid=cid+1

10. end if

11. end if

12. End for

13. End for

14.End

Points defined in the algorithm is either represent the whole points in SD or represent points of the cluster resulted from the previews iteration (Ester et al. 1996).

Algorithm 2.2: Expand Cluster(points , p, cid, Eps, Minpts): Boolean

Input: points in SD, p ∈ SD , cluster id( cid) , Eps , density threshold Minpts.

Output: return true if there is a new cluster

1.Begin

2. seed={∀ point q|q is unclassified}

3. If N_ε(q,Eps) < Minpts then

4. mark q as noise

5. return false

6. while seed is not empty

7. Begin

8. take point q from seed and mark it as classified

9. if N_ε {q, Eps} ≥ Minpts then // core point

10. q.cid=current cid

11. If any point in N_ε(q, Eps) is marked as noise