基于机器学习的电信诈骗危险预测研究

Research on the risk prediction of telecommunication fraud based on machine learning

提前检测电信欺诈行为能减少消费者的损失,有助于维护社会经济秩序。分析银行卡交易地点与家的距离等7个变量与诈骗行为的关系,分别建立Logistic回归、决策树、XGBoost等诈骗行为预测模型。研究结果表明:1)不通过芯片(银行卡)进行的交易、交易时不使用PIN码、在线交易订单均会提高被诈骗的概率;2)对比分析准确率等评价指标发现XGBoost算法在电信诈骗行为预测效果上表现更好。

The early detection of telecommunications fraud can reduce consumers losses and help maintain social and economic order.This thesis analyzes the relationship between seven variables such as the distance between the bank card transaction location and home and the fraud behavior,and establishes the fraud behavior prediction models such as Logistic regression,decision tree and XGBoost respectively.The research results show that:1)The transaction without chip(bank card),the transaction without PIN code,and the online transaction order will increase the probability of fraud;2)Comparing and analyzing the accuracy and other evaluation indicators,it is found that XGBoost algorithm performs better in the prediction of telecommunication fraud.