DBSCAN: Part 2

Hello and welcome. In this video, we'll be covering DB scan. A density-based clustering algorithm which is appropriate to use when examining spatial data. So let's get started. Most of the traditional clustering techniques such as K-Means, hierarchical, and Fuzzy clustering can be used to group data in an unsupervised way. However, when applied to tasks with arbitrary shaped clusters or clusters within clusters, traditional techniques might not be able to achieve good results that is, elements in the same cluster might not share enough similarity or the performance may be poor. Additionally, while partitioning based algorithms such asK-Means may be easy to understand and implement in practice, the algorithm has no notion of outliers that is, all points are assigned to a cluster even if they do not belong in any. In the domain of anomaly detection, this causes problems as anomalous points will be assigned to the same cluster as normal data points. The anomalous points pull the cluster centroid towards them making it harder to classify them as anomalous points. In contrast, density-based clustering locates regions ofhigh density that are separated from one another by regions of low density. Density in this context is defined as the number of points within a specified radius.A specific and very popular type of density-based clustering is DBSCAN.DBSCAN is particularly effective for taskslike class identification on a spatial context.The wonderful attributes of the DBSCAN algorithm is that it canfind out any arbitrary shaped cluster without getting effected by noise.

[@cansurmeli]

Even though the presenter has a good explanation style, this video contains crucial mistakes. First off, for a point to be border, it has to have less than M points in it's circle and(in the video, this condition is explained as `or`) be reachable by another core point in it's circle. As the video progresses, based on these conditions, some points have been falsely classified as Core or Border Points. For instanec, the points to the right in the video are actually outliers based on the given conditions of R=2, M=6.

[@MachineLearningTV]

Great explanation... Thanks

[@wayneosaur]

How is are 5 pts on the right considered a cluster when M = 6? None of them can be core points and no core points are reachable from *any* of those 5 pts.

[@DDMT_Development]

It does have mistakes i.e. calling a point a Core point when it's not, but the explanation is enough to understand the point. Thank you.

[@calvinlee6911]

This is a great and clear explanation of DBScan. However, please be responsible and make a correction post in the comment section. It’s really confusing people, much thanks!

[@tald747]

Excellent explanation, simple, short and to the point. Well done 👍

[@yahyazahlane1337]

Thank you very much for this simple explanation of DBSCAN, this is the best explanation of DBSCAN I've found so far

[@Z_Doctor]

At 5:52 the point is labeled as a core point despite there only being a total of 5 points in that cluster. I did not hear a rule stating how/why it would be labeled as such. Wouldn't it be labeled as a border point? Does a cluster need to have a core point? Would a group of border points be considered outliers?

[@wayneosaur]

Yes. This demo breaks its own rules.

[@KICKinYaFACE]

What happens if a "border point" is selected in the very first step, but it is classified as an outlier, because none of the reachable points were classified as a core point yet? Does it get re-evaluated?

[@zoyeHow]

wow, makes it so easy
to understand

[@aminzaiwardak6750]

Thanks a lot, you explain very well.

[@XuanTran-ri1hn]

Thank you very much for your great video! May I ask about the minute 5:29? M=6 means that that circle should have 6 points to have that point as core point, that circle has only 5 point, so in my opinion, it should be a border point instead. Would you mind to explain more?

[@jackyhuang6034]

Now I know why IBM isn't leading AI/ML. They even get the basics wrong.

[@BogdanAnastasiei]

Excellent video! If you allow a question: how can we know which method is more appropriate for our situation: k-means or DBSCAN? Thank you!

[@gl8218]

Does the core point we pick first is included in the M from the begining?

[@185283]

Why is right one a cluster if minpoint is 6

[@leobutracio]

I agree with you. I think the points of the seconds cluster are noise ones

[@jingyeqiu609]

@@leobutracio Or maybe we should change the minpoint to 5

[@leobutracio]

​@@jingyeqiu609 Exactly, in that case, there would be 2 clusters. And some border points would be core points.

[@MoMaYNOY]

how to calculate eps if my data is latitude, longtitude
if i want eps = 200 meter how value of eps?
or you recommend what tool?

[@hARRYnhariprasathnallasamy]

but here also we have to define minimum number of points and radius. any way to handle that better.?

[@MachineLearningTV]

Exactly... The minimum number of points and the radius affect directly the shape and number of the clusters that DBSCAN finds

[@SovietNuclear1]

In the example, the right cluster only have 5 points but M=6, isnt right one become outlier?

[@MachineLearningTV]

No.. 5 is the number of neighbors. So 5 + 1 (the point itself) = 6

[@kotetsu954]

@@MachineLearningTV he mean the last core points that u mentions sir, it's totaly just 5 points (even with the point itself),

[@wayneosaur]

@@MachineLearningTV No .. it is 4 + 1 = 5 < 6.

[@tsandbox1]

5:29 how it become core point

[@jihanapriliana5067]

how to determine the parameter?

[@MachineLearningTV]

See this link: stats.stackexchange.com/questions/88872/a-routine-to-choose-eps-and-minpts-for-dbscan

[@abdullahalnoman2411]

Even though it provides a good explanation, but there are lots of mistakes in the simulation process. People should dislike the video from here on. So that CZcams stops recommending this misleading video, or someone who came here, become alert upfront, and not waste time.

[@MachineLearningTV]

It is your opinion ok? This video has helped lots of people