An Efficient Clustering Approach using DBSCAN

Author Name(s): R. V. Sravana Kumar, Dr. S. Sreenivasa Rao, Dr. P. Srinivasrao
Author Email:


With the fast development of the Internet, the information available in the database showing explosive growth where it wouldn’t be that easy to obtain desired information accurately from the available data. This unexpected growth of internet has also led to the frequent usage of e-commerce sites which concentrate on sales based on their relevant products. To do so clustering of products is done so as to identify the sales with profits and which are lagging behind by finding similarities among the products. The clustering analysis is done with the help of various clustering algorithms like DBSCAN, K-means, GRID based, Hierarchical based, etc., .Each algorithm will produce different results; you’ll never be certain whether one result is better than the other or even whether the result is of any value. Kmeans algorithm is preferable choice for datasets having small number of clusters with proportional sizes and linearly separable data and also you can scale it to use on very large data sets. K-means algorithm needs the number of clusters fixed in advance and it arises problems when the data contains outliers. Also the downside of K-means is it leads to problems when clusters are of different sizes, densities and non-globular shapes. The DBSCAN implementation does not require any user-defined initialization parameters to create an instance like Kmeans do. So DBSCAN algorithm is more advantageous when compared to other algorithms. It is because DBSCAN doesn’t need one to specify the amount of clusters within the knowledge a priori, as against K-means. DBSCAN will notice different arbitrary formed clusters. Density clustering (DBSCAN) seems to correspond more to human intuitions of clustering, rather than distance from a central clustering point (K-means).


Massive growth of data available in database is taking place with increase in development of internet and its usage[1]. There are many applications with massive amount of data which are creating limitation in data storage capacity and processing time. Considering such huge amount of data it is not possible to obtain desired information accurately from this available data. In such scenario of growing data, business models like e-commerce has been evolving rapidly. This rapid evolution of e-commerce has few requirements like: analyzing the sales, demandmanufacture ratio, establishing new business models based on market fluctuations/scenario for given goods. For example, consider in an e-commerce site we want to improve our sales by recommending relevant products to our customers[2]. In such case initially the company must need to know about the sale of their products so as to check which product is hiking in sales with good benefits and profits and which are lagging behind[3]. But this type of process wouldn’t show better results all the time. Find clusters based on the products that the users have bought and by this clusters we find similarities between customers . E-commerce business and retailers work in a competitive and fast paced environment with the online presence these days even though influenced by cost and online advertisements. For separate information sources and data needs conventional approaches were divided with individual analytical tools. Gathering the information along is often finished manually, a human operated method. Decisions are consequentially suboptimal owing to the volume, variety and velocity of information. The challenge here is not only about gathering data but also analyzing and using it appropriately. Also it can be difficult for linking, matching, correlating and interpreting the data obtained from different sources. These are the main emphasis of clustering analysis where it assures results acquired by superior data.


This article experimented on Employee and US census data Datasets using Data mining clustering technique called DBSCAN, and the Results are Evaluated using Cluster metrics such as Purity, NMI, Rand Index, F-measure, Homogeneity, Completeness, V-measure. This methods improves the Grouping quality

417 total views, 1 views today

Download PDF File

About the author: admin