I would like to know your opinion about dbscan clustering, I am trying to implement algorithm as published here. In my opinion there is possibility for one point from border of some cluster to be an core point of another one as shown in picture:
.
I think there are some of the possible solutions:
- we could consider point as written to cluster and that cannot be changed - but we could lost second cluster because of that
- we could be able to change border points cluster but without recomputing epsilon neighbourhood.
- we could be able to add point into multiple clusters (worst one).
What do you think is the best? Or am I getting something completely wrong?