Automatic Clustering Algorithm Based On Multi-objective Clustering Ensemble

With the rapid development of database and network information technology, lotsof data can be held easily. How to get useful information from these large amounts ofdata becomes an urgent problem to be solved. Data mining is a very important branch ofinformation technology development. Clustering technology in data mining is born withthe use of the information processing of large amounts of data. Clustering isunsupervised, which is more in line with reality. In many cases, we are often unable toget any priori information about the given data. Cluster analysis will be furtherdeveloped for its widely application.In this paper, a multi-objective cluster method based on a new special crossoveroperator and a multi-objective cluster ensemble method is proposed. The works of thispaper are concluded as follows:1A multi-objective cluster method based on a new special crossover operator isproposed. In this method, first, a new crossover operator is applied. Second,another new operator is proposed, which can get new partitions by searchingthe subspace of the data by applying different algorithms which can generategood solutions. In the experiments,26gene expression data sets and5UCIdata sets are tested.The new proposed algorithm is also compared to severalsimilar algorithms, and the experimental results show that it is superior toothers on getting good solutions.2A multi-objective cluster ensemble method based on four objective functionsis proposed. In this method, first, we make a further improvement on thecrossover operator of the algorithm. Second, we apply a new objectivefunction which represents the similarity between the individual to beevaluated and the others in that current generation. There are three advantagesof this function. First, it can ensure a diverse population. Second, it can savegood solutions. Last, it can lead the crossover operator to generate goodpartitions. In the experiment section,26gene expression data sets and5UCIdata sets are tested. The results show that our method gets good results formost data sets.