A smart intelligent approach based on hybrid group search and pelican optimization algorithm for data stream clustering

Big data applications generate a huge range of evolving, real-time, and high-dimensional streaming data. In many applications, data stream clustering regarding efficiency and effectiveness becomes challenging. A major issue in data mining is clustering of data streams. The several clustering techniques were implemented for stream data, but they are mostly quite restricted approaches to cluster dynamics. Generally, the data stream is an arrival of data sequence and also several factors are added in the clustering, which is rather than the classical clustering. For every data point, the stream is mostly unbounded and also the data has been estimated atleast once. It leads to higher processing time and an additional requirement on memory. In addition, the clusters in each data and their statistical property vary over time, and streams can be noisy. To address these challenges, this research work aims to implement a novel data stream clustering which is developed with a hybrid meta-heuristic model. Initially, a data stream is collected, and the micro-clusters are formed by the K-Means Clustering (KMC) technique. Then, the formation of micro-clusters, merge and sorting of the data clusters, where the cluster optimization is performed by the Hybrid Group Search Pelican Optimization (HGSPO). The main objective of the clustering is performed to maximize the accuracy through the radius, distance and similarity measures and then, the thresholds of these metrics are optimized. In the training phase, a stream of clustering threshold is fixed for each cluster. When new data comes into this stream clustering model, the output of training data is measured with new data output that is decided to forward the data into the appropriate clusters based on the assigned threshold with minimum similarity. Through the performance analysis and the attained results, the clustering quality of the recommended system is ensured regarding standard performance metrics by estimating with various clustering and heuristic algorithms.