The core platform of intelligent risk control risk control decision engine (a)

Editor:

The rise of Internet banking, finance and science and technology to traditional financial penetration, intelligent risk control platform came into being.

The decision engine as the central role of intelligent risk control platform, the Internet is essential in the financial wave of contemporary, before the introduction decision engine, we must first understand what is the big risk control data.

First, what is big data risk control?

Baidu Encyclopedia explained: Big Data Big Data that is risk control risk control is the risk control and risk warning to the borrower by the method of using big data to build the model.

The abstract is:

Risk control decision engine as a carrier model, in fact, is to realize big data tools of risk control.

Second, what is the risk control decision engine?

Wind control decision engine for complex business logic abstract business rules stripped out combinations of different branches, association, and operation progressive layers of rules, the decision result of the final output product.

  1. Traditional wind control logic judgment decision engine to achieve the main rules, for example: when the ladies room of the rules could be developed into a "sex for women, to enter, or can not enter," people in the gender and therefore the data segment input is "male", then It determines rule can not enter;
  2. Existing risk control decision engine commonly used in the traditional basis of more feature-rich, can achieve a variety of types of rules, scorecard model, nested logic expressions, etc., to achieve a richer level of logical operations to meet today Internet financial business requirements;
  3. High-end risk control decision engine is integrated into the self-made language processing platform, stream computing platform in the existing risk control decision engine, improved calculation power and deal with the aging of existing decision engine;

Now mainly introduce risk control decision engine platform normally used, common function module contains mainly rules, scorecard model, expression, decision-making flow.

Third, the rules module

Common rules module products mainly include implementation rule sets, rule table, the rule tree.

1. Rule Set

Wherein the rule set is divided into general rule, the rotation principle, the general rule variables, expressions, conditional values, the decision result of the composition as follows:

Variables: Age, said the expression: greater than or equal, the condition value: 18, this is just a rule of the rule set, and in which there is, or logical relationship between the rules and the rules, then that is the result of the decision: meet rule1, an output member name The name "gold Member", does not meet the output member name "ordinary members."

Cyclic rules can be set on the object of rules cycle, a cycle rule may have one or more cyclic units, each circulation unit is a general rule, the rule defined in the same way as normal.

只是在执行的循环规则时,需要添加循环条件,以及循环结束后输出的决策结果,在风控决策引擎中,循环规则运用的较少,这里不做详细的讲解,感兴趣的可以留言讨论。

2. 规则表

规则表是一种表格形式的规则工具,在处理判断条件较多的时候,决策结果较多的情况时,可以快速定义出决策规则。

规则表分为条件列、决策列,其中上图借款人年龄、借款人是否有驾照、借款人命中黑名单是条件列,决策结果是决策列。

现在虽然风控决策结果输出的结果类型不要求多样化,但是规则种类、数量很多,采用规则表方案实现规则的决策配置可以更加便捷、清晰。

4. 规则树

规则树也是规则集的另一种表现形式,在展示上更加形象,在风控业务上通过规则树、规则表进行规则的配置可以更加形象、快捷。

其中每条规则的实现方式同普通规则,都有变量、表达式(条件)、条件值、决策结果(变量赋值)构成。

四、评分卡模块

评分卡是对目标的信息进行分析打分的表达方式,表示此人或此机构由于信用活动的拒付行为所造成损失风险的可能性,评分通常用于对个人或机构的风险管理与评估。

评分卡实际也是规则的变形,通过有变量、表达式、条件值、得分四部分组成,当然评分卡还会有得分的计算方式,例如求和、加权求和等。

五、模型模块

通过主观意识借助实体或者虚拟表现构成客观阐述形态结构的一种表达目的的物件(物件并不等于物体,不局限于实体与虚拟、不限于平面与立体),风控决策引擎中使用的模型更多的是数据模型,描述的是目标的行为和特征。

模型在决策引擎中,对于决策引擎平台实际是一个已经封装好了的产品,决策引擎只会负责入参变量的配置、出参变量的配置以及模型的调用,所以这个模块的核心主要是考虑模型的类型(py、model)、调用逻辑、入参以及出参变量的配置。

六、表达式模块

表达式模块主要是规则、评分卡等逻辑判断实现困难时,可以直接通过代码自由编辑实现决策的规则判断,其中规则的表达式、条件值、决策结果都是通过编码实现,通过这样的方式可以运用于更多小众难实现的决策场景,灵活性更大。

表达式模块类似模型模块,规则的入参和出参配置也是重点。

七、决策流模块

决策流它实现整个分开工决策引擎的工作流配置,用来对已有的规则、评分卡、模型、表达式进行执行顺序的编排,清晰直观的实现大型、复杂的风控规则。

决策流核心的构成包含“开始节点、规则/评分卡/模型等已封装好的规则包节点、决策节点、分支节点、聚合节点。

  1. 开始节点为一个决策流开始的地方,决策流程必须有始有终且必须以开始节点作为开始;
  2. 规则包节点,实际就是用来添加之前在规则、评分卡、模型、表达式中已经创建好的规则产品;
  3. 决策节点是在决策时,根据为其下流出连接配置的条件来决定究竟应该走哪条连接的节点,所以根据这一特性,决策节点下流出连接至少要有两条,否则决策节点就没有意义了;
  4. 分支节点实现规则流多条并行的节点,通过这个节点,可以根据当前节点下流出连线数量,将当前规则流实现拆分成若干条子的规则流实例并行运行;
  5. 聚合节点用来聚合由分支节点拆分出来的多个子的规则流,实现多条规则流的汇合;

有始有终,决策流程的结束,一般是伴随着决策总、分的流程的执行,执行到最后节点自动结束,输出决策结果。

决策引擎除了以上核心功能模块以外,实际上为了风控决策引擎灵活多变,能够实现尽可能多的风控业务场景,通常会实现规则、评分卡、表达的相互嵌套调用,这样可以更好应对不同的风控业务场景。

以上只是对风控决策引擎做了简要的介绍,其中的规则、评分卡等功能在风控业务复杂的情况下还可以对规则和评分卡进行产品升级,实现复杂规则、复杂评分卡的决策能力。

实际应用中的产品只靠风控决策引擎是远远不够的,风控决策引擎的应用还会搭配指标平台、接口管理平台、风控报告等产品一同服务于风控业务。

关于复杂规则、复杂评分卡、决策引擎配套产品欢迎讨论,后期我会为大家逐一呈现。

 

作者:互金杂货铺,微信号:zjlove778。

本文由 @互金杂货铺 原创发布于人人都是产品经理。未经许可,禁止转载。

题图来自unsplash,基于CC0协议

Guess you like

Origin www.cnblogs.com/maohuidong/p/12166685.html