朴素贝叶斯小样本金融客户分类方法与分类偏好研究

对金融客户进行准确分类是向其提供个性化服务的前提.针对某金融产品的销售需求,通过在线推销测试收集客户样本数据,并根据用户反馈标注样本.通过构造概率分布函数、离散化连续型数据两种方式构建贝叶斯分类器.利用交叉检验训练和测试分类算法,发现朴素贝叶斯分类算法性能优于高斯贝叶斯算法和逻辑回归算法.离散化连续型数据过程中结合分类偏好进行数据过滤,实验证明,异常数据滤除率参数对客户分类算法的准确性有显著影响,通过恰当设置该参数的取值,可以调节分类算法的分类偏好.方法对于提升金融产品销售效率,降低营销成本有参考价值.

Preference Classifier of Few-shot Finance Clients Based on Naive Bayesian

Classification of finance clients is prerequisite to provide individual service to them.In order to sell one financial production,clients information is collected through experimental online sale.Clients are marked according to their reaction to the financial production.Two Naive Bayesian classifiers are built by two methods.One is to construct probability distribution functions on clients sample data.The other is to discrete clients sample data.These classifiers are fitted and tested using cross validation on these finance clients data.The performance of Naive Bayesian classifier based on data discretization is the better than Gaussian Naive Bayesian classifier and Logical Regression classifier.In the process of data discretization some data is filtered according to classifier preference.Experiment show the ratio of filtering alternative data apart from the main data has an important influence on classifier accuracy.Classifier preferrence can be adjusted by setting appropriate value to the ratio.Naive Bayesian classifier based on data discretization is useful to promote finance production sale.