DBSCAN stands for Density-Based Spatial Clustering of Applications with Noise.
It is a popular unsupervised learning method used for model construction and machine learning algorithms. It is a clustering method utilized for separating high-density clusters from low-density clusters. It divides the data points into many groups so that points lying in the same group will have the same properties. It was proposed by Martin Ester, Hans-Peter Kriegel, Jorg Sander, and Xiaowei Xu in 1996.
DBSCAN is designed for use with databases that can accelerate region queries. It can not cluster data sets with large differences in their densities.
- It identifies clusters of any shape in a data set, it means it can detect arbitrarily shaped clusters.
- It is based on intuitive notions of clusters and noise.
- It is very robust in detection of outliers in data set
- It requires only two points which are very insensitive to the order of occurrence of the points in data set
- Specification of number of clusters of data in the data set is not required.
- It can find any shape cluster even if the cluster is surrounded by any other cluster.
- It can easily find outliers in data set.
- It is not much sensitive to noise, it means it is noise tolerant.
- It is the second most used clustering method after K-means.
- The quality of the result depends on the distance measure used in the regionQuery function.
- Border points may go in any cluster depending on the processing order so it is not completely deterministic.
- It can be expensive when cost of computation of nearest neighbor is high.
- It can be slow in execution for higher dimension.
- Adaptability of variation in local density is less.