PreFIRST [Predictive Fraud Identification and Reduction Statistical Technology]

汽车欺诈是英国犯罪的主要来源，随着电子营销和在线服务的发展，欺诈的机会正在迅速增加。根据英国国家反欺诈局的统计数据，英国每年的汽车欺诈金额估计为7.7亿英镑。车辆诈骗的目标是以没有现金或现金极低的价格（例如支票诈骗、假买家、存款诈骗）或以高价出售低价值车辆（例如里程诈骗、克隆车辆、不存在的车辆）来赎回消费者的车辆。随着越来越多的交易转移到网上，制造数字库存（看起来真货的假网站，拥有“真”凭证的假汽车）变得更加容易。目前的汽车检查服务依赖于历史数据（被盗、财务、事故等）。没有可用的服务为潜在的欺诈提供警报或标记。这一概念验证的目的是展示从通过历史检查量化汽车数据到准确预测欺诈企图的一个步骤变化。该项目名为“PreFIRST”，即预测欺诈识别和减少统计技术，重点关注欺诈的早期识别。这个概念是一个集成的欺诈管理平台，包括实时数据监控和复杂的行为检测，包括来自Facebook和Twitter的社交网络信号和典型用户的语义。风险监测和检测分析的创新组合，旨在机器学习和预测欺诈模式、特征和行为，将为直观的“风险评分”工具提供信息，提醒利益相关者并提供有效的决策支持，以打击欺诈。访问多个实时数据源将提供更广泛的“大数据”视图，以提高预测的准确性。对这些来源的数据进行分析并转换为“标准化”格式，使PreFIRST能够利用数据挖掘算法和知识发现来评估和预测潜在欺诈事件的可能性

Automotive Fraud is a major source of crime in the UK and opportunities for fraud are increasing rapidly with the growth in e-marketing and online services. Based on National Fraud Authority statistics the estimate for automotive Fraud is £770 million per year. Vehicle fraud is targeted at relieving a consumer of their vehicle for no or very low cash (e.g. cheque fraud, fake buyers, deposit scams) or selling a low value vehicle for a high price (e.g. mileage scams, cloned vehicles, non-existent vehicles). As more transactions move online, it has become easier to produce digital inventory (fake websites that look genuine, fake cars with ‘genuine’ credentials). Current car checking services rely on historical data (stolen, on finance, accident etc). There is no available service that provides an alert or flag for potential fraud.The objective of this Proof of Concept is to demonstrate a step change from quantifying car data via history checks to accurately predicting attempts to defraud.The Project - “PreFIRST” - Predictive Fraud Identification and Reduction Statistical Technology - focuses on early identification of fraud. The concept is an integrated fraud management platform encompassing real time data monitoring with sophisticated behaviour detection, including social network signals from Facebook and Twitter and semantics for typical users. An innovative combination of risk monitoring and detection analytics, designedto machine learn and predict fraudulent patterns, traits and behaviours will feed into an intuitive ‘risk scoring’ tool that will alert stakeholders and provide effective decision support to combat fraud.Access to multiple live data sources will give a wider ‘big data’ view to increase the accuracy of prediction. Data from these sources is analysed and transformed into a ‘standardised’ format to allow PreFIRST to utilise data mining algorithms and knowledge discovery to evaluate and predict the likelihood of a potential fraud event