DynamicEvoStream: An EvoStream based Algorithm for Dynamically Determining The Number of Clusters in Data Streams

EvoStream is a stream clustering algorithm which gradually clusters data in the idle times of the stream. In comparison with other algorithms in this field, EvoStream has a lower computation overload in the offline phase and has better accuracy. Also, in this algorithm, the number of clusters is taken as constant whereas in an authentic stream this number varies with the complexity of input data. In this work, we present DynamicEvoStream as an improved version of the original EvoStream. In this algorithm, we detect and exploit variations in the distribution and speed of the stream.  Also, we modified the cleanup function to merge overlapping clusters. Therefore, in contrast to the basic EvoStream, DynamicEvoStream identifies the number of clusters in a dynamic manner. Also, the speed of evolutionary steps is increased while improving the quality of the clusters. Finally, experiments using DynamicEvoStream on different streams showed that it can cluster the stream up to four times faster than the original EvoStream with fewer computation and memory resources. In the worst case, the quality of clusters is competitive to the original EvoStream, however improves the quality of clusters up to 30% in the majority of cases.