improving the effectiveness of the service as time goes on. By adopting a data-driven approach to banking, these mobile banks can offer features that established banks have yet to provide, such as the ability to set up an account in minutes; instant card locking and unlocking via the app, and real-time tracking of spending through app notifications and smart spending insights. These features have already proven extremely popular, so much so that both Monzo and Starling have won awards for 'Best Banking App' and 'Best British Bank' respectively [3].
然而，在过去几年中，数字银行的兴起和挑战者银行的形成使得银行业消费者生成的数据得到了充分利用。由于最近引入了开放式银行业改革，要求银行允许您与授权提供商共享您的财务信息[1]，因此“Monzo”或“Starling Bank”等纯粹基于应用程序的银行的创建得到了促进。这类银行没有任何分行，可以通过应用程序中的实时聊天提供几乎全天候的即时支持，这是传统银行难以提供的。这些实时聊天使用机器学习来检查聊天中客户发送的内容，并为客户服务代理生成可能的适当响应。通过扫描和分析书面查询中的语言，可以加快提交请求的过程，有助于连接用户并更有效地帮助团队工作人员[2]。此外，机器学习算法将了解什么响应最适用于给定的请求，随着时间的推移，进一步提高服务的有效性。通过采用数据驱动的银行业务方式，这些移动银行可以提供老牌银行尚未提供的功能，如在几分钟内建立账户的能力；通过应用程序即时锁定和解锁卡，并通过应用程序通知和智能消费洞察实时跟踪消费。这些功能已经证明非常受欢迎，以至于Monzo和Starling分别获得了“最佳银行应用程序”和“最佳英国银行”奖[3]。
App-based banks are examples of a type of company referred to by the umbrella term 'FinTech', which describes the evolving intersection of financial services and technology [4]. It also often refers to the technologies themselves that are having a disruptive influence on existing financial process. Although in the early stages of its life, FinTech has already had a dramatic impact on the global economy. In 2019, 64% of FinTech service consumers worldwide have previously used at least one platform, up from 33% in 2017. There has also been an effect on start-up companies with 25% of global SMEs (small and medium-sized enterprises) have adopted fintech services for use in banking, financing, and financial management [5, see EY Global FinTech Adoption report].
基于应用程序的银行是“FinTech”这一总称所指的一类公司的例子，它描述了金融服务和技术不断发展的交叉点[4]。它还常常指对现有金融流程产生破坏性影响的技术本身。尽管在其生命的早期阶段，FinTech已经对全球经济产生了巨大的影响。2019年，全球64%的FinTech服务消费者此前至少使用过一个平台，高于2017年的33%。全球25%的中小企业采用了金融科技服务，用于银行、金融和金融管理[5，见安永全球金融科技采用报告]，这也对初创公司产生了影响。
Fraud Detection 欺诈检测
Another key area in which data science has transformed the finance industry is fraud detection. Fraud in the financial world can take many different forms, from improper payments and money laundering to terrorist financing and cybersecurity breaches. In the past, fraud detection used defined business rules and rudimentary analytics in order to search for anomalies. This meant that those investigating potential cases of fraud often had to do so after it had been identified and a crime was committed - it was a reactive approach rather than a proactive one. Nowadays, newer technologies and the development of the data science field has led to a complex and sophisticated set of solutions and approaches that have revolutionised fraud detection and prevention. By moving away from standard analysis and adopting predictive and adaptive techniques; using the vast amounts of transactional data society now produces on a daily basis; and employing realtime monitoring and analytics, the tide has started to turn in favour of prevention over reaction.
数据科学改变金融业的另一个关键领域是欺诈检测。金融世界中的欺诈可以采取许多不同的形式，从不当支付和洗钱到恐怖融资和网络安全漏洞。在过去，欺诈检测使用定义的业务规则和基本分析来搜索异常。这意味着，那些调查潜在欺诈案件的人往往必须在发现欺诈案件并实施犯罪后才这样做——这是一种被动的做法，而不是主动的做法。如今，新技术和数据科学领域的发展带来了一套复杂而复杂的解决方案和方法，彻底改变了欺诈检测和预防。远离标准分析，采用预测和自适应技术；使用当今社会每天产生的大量事务性数据；通过实时监控和分析，趋势开始转向预防而非反应。
Fraud detection in a machine learning context is often framed as a classification problem [6]. This is where we try to classify each data point to a certain class label based on the values of the attributes of that instance. For a given output of the algorithm, each instance is assigned to a class and can be plotted to show a visualisation of the different regions corresponding to each class label (see Fig.1).
机器学习环境中的欺诈检测通常被视为一个分类问题[6]。在这里，我们尝试根据实例属性的值将每个数据点分类到某个类标签。对于算法的给定输出，每个实例被分配到一个类，并且可以绘制以显示对应于每个类标签的不同区域的可视化（参见图1）。
Figure 1: An arbitrary example of a classification problem in machine learning for two class labels. 图1：机器学
