| With the development of science and technology,it has brought a lot of convenience to people.Now people can purchase their favorite products online through mobile devices without leaving their homes.However,bringing convenience to people,at the same time it also puts us in an era of information explosion..It is difficult for us to find valuable information form data.People usually use cluster analysis to divide the data,which greatly reduces the time we search information.The K-means algorithm is one of the more common ones,but the K-means algorithm deals with some special data.Sometimes it will expose its shortcomings.In view of the shortcomings of the k-means algorithm,the aim of the paper is to improve the k-means algorithm from three aspects:(1)The traditional k-means algorithm uses the Euclidean distance to calculate the difference between data,and(2)the k-value fuzziness and subjectivity of the k-means algorithm.(3)The k-means algorithm for calculating the cluster center does not measure the relationship between classes and classes,it is affected by noise data and performance is not very good.In this paper,the improved k-means algorithm will be used to classify the data.The improved K-means algorithm will use the coefficient of variation to determine the weight of each coordinate,and the weight of the data will be verified by the way people agree.The above coefficient of variation method is used to calculate the feasibility of the distance between data.The k value is determined by the functional relationship.By calculating the distance between classes,it is then determined which classes are put together for the next cluster center calculation.Using the method of variation coefficient to calculate the distance between the data,it can more embody the emphasis on different goods.The coefficient of variation will calculate the weight of each variable of the different commodity by scientific calculation,so that the product can reach the best result in the cluster.The K value is confirmed by the function relation,which is the distance sum from the classmembers to the respective center points and each cluster center to all class member centers(value of all class members as the class member center).When the value of the function is taken to the minimum,the clustering is the most reasonable,and the K value at this time is the ideal K value.In order to optimize the performance of the center point,it is mainly to consider the possibility that there may be a large difference between class and class.Then,the classes with larger differences should not be put together to calculate the next center point,because the difference is large and their center points will only change in their class itself,and the advantage of this is that it significantly reduces the impact of noise data on normal data. |
