GA-DBSCAN:A New Genetic based Clustering Algorithm

In the machine learning process, different approaches are adopted to solve different problems according to the problem and application. One of the approaches that is selected is based on educational examples. Samples are divided into two categories, labeled and unlabeled. Unlabeled examples of unsupervised learning methods used that ambiguity and complexity increases. Methods available in this area include clustering methods based on partitioning, hierarchy, density, and so on. In density-based clustering, using neighborhood concepts and density defines a cluster. The input parameters of the algorithm are the minimum distance between the two points that indicate whether they are neighbors or less, and the smaller number of accessible points indicates the density of at least one cluster. These parameters strongly affect the clustering results. For this reason, one of the main challenges is choosing the right two. In this study, an attempt has been made to select the value of these two parameters optimally and to achieve this goal, Genetic algorithm has been used. The outputs generated by the genetic algorithm have optimal values and these values are used as input to the DbScan algorithm. The proposed algorithm is tested in ۴ UCI datasets and compared with the original DbScan method in terms of clustering time, accuracy and noise point detection. And compared to the original DbScan method, it is found to be more capable.

Keywords:

Unsupervised Learning , Clustering , Density-Based Genetic Algorithm