Parallel density clustering algorithm based on MapReduce and optimized cuckoo algorithm

Abstract

In the process of parallel density clustering, the boundary points of clusters with different densities are blurred and there is data noise, which affects the clustering performance and makes the clustering results subject to the influence of local optimality. A parallel density clustering algorithm based on MapReduce and optimized cuckoo algorithm is proposed to solve the problem. Firstly, the algorithm combines the nearest neighbor and inverse nearest neighbor strategies in -means (DBSCAN), and redefines the expansion conditions of clusters in the density-based spatial dutering of apps with noise (DBSCAN) algorithm by calculating the influence space of each data point, avoiding the problem of fuzzy division of boundary points of clusters with different densities; secondly, a feasible iterative noise point processing strategy is proposed by combining the nearest neighbor idea in KDBSCAN density clustering, which reduces the impact of noise points in the data on the performance of clustering algorithms; thirdly, an optimization and improvement strategy MCS (Majorization cuckoo search) based on the traditional cuckoo algorithm is proposed, which attenuates the weight of the probability of finding the nest and improves the algorithm convergence speed as the number of iterative searches increases, solving the problem that the clustering results are constrained by local optimality; finally, a parallel density clustering strategy MCS-KDBSCAN is proposed in combination with MapReduce, which reduces the communication burden of transmitting the local optimal solution of the parallel clustering algorithm by parallelizing the density clustering algorithm operation and improves the algorithm performance. Experiments show that the proposed MCS-KDBSCAN parallel density clustering algorithm is superior in terms of clustering accuracy and clustering running time.