Using Partitioned-Based Method for Optimal Epsilon Parameter Extraction on Density-Based Clustering
Hafsa Abdjalil Mansori, Omar M. Sallabi, Abdelsalam M. Maatuk · 2021 IEEE 1st International Maghreb Meeting of the Conference on Sciences and Techniques of Automatic Control and Computer Engineering MI-STA · 2021
Clustering is vital and striking in data mining, a powerful tool to analyze an incredible volume of data created by the progress in applications. The DBSCAN clustering is a well-known successful algorithm, which has features that are dependent on the density notion of objects within clusters. It can find clusters of the highest dimensionality as long as high dense regions exist, and find clusters with different shapes. Besides, it is robust in dealing with noise and large spatial datasets, so that it is important and succeeded in dealing with large datasets, i.e., big data. This paper aims at proposing an enhanced version of the algorithm, called eDBSCAN, to gain the suitable input parameter (Eps) on the classical DBSCAN algorithm using the partitioning method. We used the MC-means algorithm in the extraction of the optimal Eps value. The experimental study proves that the eDBSCAN algorithm is efficient in determining the Eps value, through a variety density included in a data set. The results achieved in purity indicate the Eps value effectiveness and accuracy to 100% percent on the data sets used.