Performance Enhancement of DBSCAN Density-Based Clustering Algorithm in Data Mining

In order to facilitate better data analysis, clustering is a prominent data mining approach which separates data into separate categories based on their similarities and distinctions. It’s a method of unsupervised learning which may generate a model with a training dataset. Throughout this study, the Euclidian distance and EPS (Epsilon) value has been determined using the DBSCAN technique based on the similarities as well as disparity for the input data. Moreover, Euclidian distance is dynamically determined using the back propagation method; a simulation study is carried out that demonstrates both speed and precision improvements. DBSCAN, or spatial clustering based on density of application with noise, is a common clustering technique for geographic datasets that can locate clusters of any shape and automatically detect noise points. To deal with the problem, an improved DBSCAN technique based on the selection of features (FS-DBSCAN) has been suggested. However, the algorithm proved unstable while handling huge dimensional data. Many simulation via real-world datasets testify the algorithm’s efficiency. Comparison of the proposed algorithm to the DBSCAN algorithm demonstrates its benefits. Structures within the collections may be identified by dividing of collections of products into classes and clusters, with a focus on increasing the resemblances of objects within each cluster and lowering the similarities of things belonging to different clusters.