Clustering Algorithm Based On Multi-task Evolution

Because of the improvement of computer computing power,the computer application technology related to data has achieved rapid development.In the era of big data,massive data is stored in the database.People's focus has shifted from obtaining and storing large amounts of data to how to mine data to obtain useful information.Scientific experimental data,medical data,demographic data,financial data and other fields have produced a lot of data mining related application technology,the development of data mining related algorithms has become an urgent demand.Data mining technology can be divided into many types according to different purposes,and one of the important branches is cluster analysis.In order to optimize the clustering mode that meets the different distribution characteristics,various clustering distribution indexes are proposed.The division evaluation of different data sets by different cluster internal indicators is not completely consistent,and only one optimization task can be completed when a single indicator is used as the optimization target.In response to this characteristic,some scholars have proposed a multi-object clustering algorithm,which covers different data sets and different target values by obtaining a large number of solutions.Multi-task learning can treat the optimization of different indicators as different tasks,find the optimal solutions under each task,and finally use expert knowledge to find the clustering result that is most suitable for the data set,and the number of solution sets obtained Far less than multi-objective algorithms.Based on the characteristics of clustering tasks,in order to improve the multi-index analysis ability of clustering optimization,this article introduced three algorithms based on multi-task optimization from the beginning of multifactorial evolution algorithm.The advantage of the clustering algorithm under the multi-task evolution framework is that it optimizes multiple clustering internal indicators at the same time,and can obtain a larger number of solution sets than the single-task optimization algorithm,and the result selection is easier than the multi-objective clustering algorithm.Finally,The target solution that is most suitable for the data set is selected through expert knowledge.The main contents of this article include:(1)This article introduces a clustering algorithm using multi-task evolutionary which is based on the framework of multifactorial evolution algorithm,and the clustering optimization ability of the algorithm in artificial and real data is tested;(2)The clustering algorithms of multi-task evolutionary and level-based multi task cooperation are introduced which are based on an automatic clustering optimization algorithm;(3)Using synthetic data sets and real data sets to test the algorithm in this paper,the multi-task algorithm shows good results in the test results.In addition,a number of experiments on algorithm parameters have been carried out on the algorithm proposed in this paper.