Improvement On K-means Clustering Algorithm And Its Application Research In Garment Production

Cluster analysis, an important technology in data mining, is based on the similarity todivide the data set into groups, which has been widely applied in statistical analysis,pattern recognition and image processing and other fields. The K-means algorithm, widelyused in the cluster analysis with the advantages of simplicity and efficiency, which,however, has such defects: K-means Algorithm is extremely dependent on and sensitive tothe choice of the initial cluster centers and it needs a large number of loop iterations toassign objects to the nearest cluster. Therefore, in this thesis, we will make some in-depthanalysis and researches on K-means algorithm, propose and design two improvedalgorithms, and at the end, we will do some application researches on the two aspects ofgarment production based on the school-enterprise cooperation project. The main task willbe as follows:(1) Against the disadvantages of the K-means algorithm, we put forward a method toinitialize cluster on the basis of sorting-partition in terms of choosing initial cluster centersand an optimization method able to assign objects to the nearest cluster in terms of loopiterations. Experiments show that such improved algorithm proves to be a simple andefficient one in processing low-dimension data, increasing the accuracy and efficiency ofcluster to a large extent.(2) Against the declining performance of K-means algorithm in processinghigh-dimension data, we come up with an improved algorithm based on the k-means ofKd-tree. In respect of initializing cluster centers, the method of partitioning subsamples byusing Kd-tree is adopted. And as for loop iterations, Kd-tree is created in cluster centers toassign objects to the nearest neighbors. Experiments show that this improved algorithm isof relatively high accuracy and efficiency and is obviously superior to traditional k-meansalgorithm in cluster accuracy. (3) Based on the school-enterprise cooperation project， Exploratory applicationresearches are made on the two aspects of garment production on the basis of clusteranalysis. One is to cluster workers and to allocate appropriate production line by using theimproved algorithm based on sorting-partition in accordance with varied proficiency inoperation. The other is to do some cluster analysis on a large amount of production hoursby adopting the improved algorithm based on Kd-tree and to formulate standard workinghours for garment production on these grounds.