TSKPSO: Spark-Based Multiple Kernel Particle Swarm Optimization Algorithm for Big Data Clustering

As data continues to grow exponentially, the need for efficient clustering algorithms has surged. Conventional clustering methods are excessively time-consuming and computationally intensive due to the enormous data sizes. In order to overcome these issues, a scalable algorithm with good clustering accuracy is required. Here a Spark-based multiple kernel clustering algorithm is proposed that combines tangential kernel and spherical kernel. And it uses particle swarm optimization to further optimize the cluster centroids and obtain the final clusters. Apache Spark is a distributed system that provides fast processing for big data workloads. The suggested algorithm is tested on skin and localization datasets. The implementation is compared against similar algorithms based on the MapReduce framework. The results show that the TSKPSO algorithm returns a DB index value of 10.95 for localization dataset and 8.17 for skin dataset when the number of clusters are 3. Both of these values are significantly lower than those returned by the MapReduce-based algorithms.