分类回归方法优化及在消费行为分析预测中应用

发布时间：2018-06-10 08:44

本文选题：决策树 + 剪枝算法　；参考：《中国石油大学(北京)》2016年硕士论文

【摘要】：随着计算机在商业领域广泛应用,积累了海量的个人消费行为数据。研究消费行为数据,分析并预测消费者的消费需求和消费倾向,对企业制定生产和营销计划具有重要指导作用。为了预测消费者的消费行为,挖掘消费行为数据中的隐藏信息,主要进行如下工作:首先,优化分类回归决策树的生成方法,通过分析数据在属性取值上的分布特性限定属性的最大分割次数;为属性选取标准添加缺失数据惩罚因子,降低将缺失取值较多的属性选取为分割节点;同时,将离散属性的分裂标准由二分裂改为多分裂,从而获取具有较高分类预测准确率和较小规模的决策树模型。其次,针对现有决策树剪枝算法的缺点,提出新的动态结合决策树的分类误差和规模量的剪枝标准,不需要独立剪枝数据集,并综合考虑决策树的分类误差和存储规模,解决单一指标剪枝出现偏向,获得更加平衡的决策树,提高决策树模型的泛化能力。最后,将优化改进的分类回归决策树生成算法和决策树剪枝算法应用于划分消费群体和预测消费者购买倾向。
[Abstract]:With the wide application of computer in the field of business, a large amount of personal consumer behavior data has been accumulated. The study of consumer behavior data and the analysis and prediction of consumer demand and propensity play an important role in guiding enterprises to formulate production and marketing plans. In order to predict the consumer's consumption behavior and mine the hidden information in the consumer behavior data, the main work is as follows: first, optimize the method of classification and regression decision tree generation. By analyzing the distribution of data on attribute values, limiting the maximum number of attribute segmentation, adding missing data penalty factor for attribute selection criteria, reducing the missing values of the attributes selected as the split node, at the same time, The split criterion of discrete attributes is changed from two-splitting to multi-splitting to obtain a decision tree model with higher classification accuracy and smaller scale. Secondly, aiming at the shortcomings of the existing decision tree pruning algorithms, a new pruning criterion for dynamic combination of decision tree classification error and scale is proposed, which does not need independent pruning data set, and considers the classification error and storage scale of decision tree synthetically. A more balanced decision tree is obtained by solving the deviation of single index pruning, and the generalization ability of decision tree model is improved. Finally, the improved classification and regression decision tree generation algorithm and the decision tree pruning algorithm are applied to partition consumer groups and predict consumer purchase tendency.
【学位授予单位】：中国石油大学(北京)
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP311.13

【参考文献】