2

I have file with coordinates X, Y of point objects in EPSG3301 coordinate system (that means it is in meters).

I want to spatially cluster points. Here is my R code (dbscan package):

db2 <-dbscan(cbind(point2$X, point$Y), eps = 500, MinPts = 25)

In what units is eps in this case? How can I achieve eps to be axactly 500 meters?

1

There are many packages for computing DBSCAN:

Which do you use ?

The unit of EPSG3301 is meter.

From Wiki Books: Data Mining Algorithms In R/Clustering/Density-Based Clustering

The elements of the database can be classified in two different types: the border points, the points located on the extremities of the cluster, and the core points, which are located on its inner region. A neighborhood of a point p is a set of all points that have a distance measure less than a predetermined value, called Eps

In your case the distance measure is therefore in meters (euclidean)

0

You do not seem to be trying the relevant function arguments to test if the eps argument is defined as euclidean distances. That said, it may help to look at the help for the fpc::dbscan implementation as the help is a bit more informative and the model specification/parameters almost identical. The dbscan package implementation is just an optimized version of the fpc version. However, keep in mind that the two model parameters "eps" and "minPts" interact in a way that may not result in an "exact" search distance.

The require input for dbscan::dbscan specifically states a matrix that can be a distance object. In other terms, a matrix of coordinate pairs or actual distance matrix. You can have an underlying assumption that a matrix of projected coordinate pairs will be calculated using euclidean distances.

From dbscan::dbscan help:

x - a data matrix or a dist object.

search - nearest neighbor search strategy (one of "kdtree" or "linear", "dist").

From fpc::dbscan help:

data - matrix, data.frame, dissimilarity matrix or dist-object. Specify method="dist" if the data should be interpreted as dissimilarity matrix or object. Otherwise Euclidean distances will be used.

method (same as search) - "dist" treats data as distance matrix (relatively fast but memory expensive), "raw" treats data as raw data and avoids calculating a distance matrix (saves memory but may be slow), "hybrid" expects also raw data, but calculates partial distance matrices (very fast with moderate memory requirements).

The search/method argument then should probably be "dist" to ensure that euclidean distances are used. However, you should also perform some exploratory analysis to ensure that this is a supported value for eps. Take a look at dbscan::kNNdist

In reference to @gene, you should be aware that the dbscan::dbscan function supersedes the implementation in the fpc package.

Note: use dbscan::dbscan to call this implementation when you also use package fpc.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.