An Evolutionary Stream Clustering Technique for Outlier Detection

Clustering data streams appeared to be the most popular studies among the researchers due to their developing field. Data streams address numerous threats on clustering such as limited time, memory and single scan clustering. Besides, identifying arbitrary shapes clusters approach are very significant in data streams applications. Data streams are an infinite sequence of the element, evolve over time with no knowledge on the number of the clusters. Various factors such as some noise appear occasionally have the potential to negatively impact on data streams environment. The density-based technique is proven to be an astounding method in clustering data streams. It is computationally efficient to yield arbitrary shape clusters and detect noise immediately. Generally, it does not require the number of clusters in advance. Most of the traditional density-based clustering is not applicable in data streams due to its own characteristics. Nearly all traditional density-based clustering algorithms can be extended to the latest ones for data streams study purposes. This concept is mainly focused on the density-based technique in the clustering process to overcome the constraint from data streams nature. This paper proposes a preliminary result on a density-based algorithm (evoStream) for clustering which is to investigate outlier detection on three different real data sets named, KDDCup99, sensor and power supply. Later, this algorithm will be extended to optimize the model in detecting outlier on data streams. © 2020 IEEE.