摘要
针对某电商旅行社平台信用卡相关历史交易数据,基于数据挖掘技术,采用多特征字段融合,在信用卡交易记录中挖掘其隐藏的、潜在的异常交易模式,从而给相关管理部门提供必要的风控预警,提取有关142个特征字段构成数据集,其中正负样本共计10110条,对原始特征映射后进行二进制非对称编码,特征维数达到801个,采用特征权重计算来抽取主要特征,对数据存在的大量冗余无关特征进行降维压缩处理,特征维数降至25个,减少了数据在预处理和模型训练过程中的内存压力,使用K-means和SVM对其进行分类学习,在真实数据集上对比实验分类结果,表明整个处理过程所采用的方法是有效的。
Based on data mining technology,we analyzed the credit card-related historical trading data of a certain e-commerce platform,dug out the hidden abnormal trading behaviors from the trading records using multi-features merging method,so as to provide necessary early warning for relevant departments. We extracted 142 features to form the sample data sets( 10110 samples in total),and obtained 801 feature dimensions through mapping and encoding. Then,using feature reduction,we decreased dimension number to 25. We then applied K-means and SVM to the classification training,and compared the results with real data sets. The results showed that the method was effective.
出处
《贵州科学》
2017年第6期83-87,共5页
Guizhou Science