Improvement Of Classification And Regression Method And Its Application In Analysis And Prediction Of Consumer Behavior

With the grand computer application in business area,massive data of consumption behavior has been collected.The research on consumption data,analysis and prediction of consumer demand and consumer inclination are of great significance for the manufacturing and marketing.In order to predict consumption behaviors and mine the underlying information in the customer behavior data,this thesis includes some work as follows:Firstly,optimize the generation method of decision tree,define the maximum segmentation times through analyzing the distribution characteristics of data.Add penalty factors of the missing data for the attributes standard selection,reduce situations that attributes are selected as segmentation nodes.Meanwhile,change the segmentation standard of discrete attribute from binary partition to multiple partition,and then acquire a relatively small decision tree model with higher forecasting accuracy.Secondly,on the basis of the current shortcomings of decision tree pruning calculation,put forward new classification error and pruning standard with new dynamic decision tree,which does not require separate data set for pruning.Besides,taking classification error and storage size into comprehensive consideration,solve problem of deviation with single index-pruning so as to gain a more balanced result from decision tree,therefore,significantly improve generalization ability of the decision tree model.Finally,apply the improved decision tree generation algorithms with the decision tree pruning algorithm to the analysis and forecast in consuming behavior and the division of consumer groups to predict the tendency of customers.