I have presented a local outlier detection method based on a scalable density-based clustering approach in which the main algorithm utilized for the clustering sections is DBSCAN. But the point is that in this method, the input data is considered as having only one unique distribution. For the various distributions case, I should follow another density-based approach, probably another enhanced version of DBSCAN which can handle datasets with different kinds of data propagation.

For this matter, I searched for various types of proposed DBSCAN enhancements that claim on successfully and efficiently clustering non-uniquely distributed data. However, I could not find any useful method which claims on exactly locating the optimal parameters Eps and MinPts, differently and specifically for every individual cluster in data.

Well, I am still searching for it, but if anyone is there that knows a perfect or semi-perfect approach that can provide such a requirement, I would be so glad to be guided.

Best of luck ...