Clustering Algorithm for IoT Data Stream Based on K-Dimensional Tree and Self-Organizing Density

Abstract

With the development of IoT technologies, hundreds of millions of devices are constantly generating sensory data streams that contain a wealth of knowledge.To derive interoperable information from them, effective methods and techniques are needed to process and analyze the data streams.The stream clustering techniques in machine learning have gained increasing attention for its ability to rapidly discover knowledge and extract insights from data streams.In this paper, an IoT data stream clustering algorithm based on K-Dimensional tree and Self-Organizing density (KDSO) is proposed.The algorithm creates new clusters using KD trees to reduce the number of redundant clusters and performs range search quickly.In addition, it follows the idea of competitive learning to absorb new data points to facilitate the merging of micro-clusters.Meanwhile, it dynamically adjusts the clustering parameters for micro-cluster update and evolution.Experimental comparisons are made with other advanced methods.The results show that KDSO outperforms the compared methods in terms of clustering purity and silhouette coefficient, and shortens the clustering processing time, proving its good clustering performance.