Prescriptive Analytics in Service Systems

业务分析，即使用数据驱动的方法来支持决策，在组织中发挥着核心作用。从侧重于以图形方式描述和统计检测现有系统中的问题的描述性分析，到旨在预测系统未来行为的预测性分析，再到以系统改进为目标的说明性分析，业务分析成为系统和流程分析和优化的支柱。在拟议的研究中，我将重点介绍服务系统中的规范分析，例如医院、公共交通系统和数据中心。我将针对诸如“如果我们在昨晚的值班期间增加5名护士，急诊科将如何表现？”，“如果早5分钟调度，公交车什么时候到达目的地？”，以及“我们如何以最佳方式安排数据中心以实现预定义的服务级别协议？”在任何组织中，尤其是在服务中，以数据驱动的方式回答这些问题是基于证据的分析和优化的最终目标。在现代，大多数服务系统使用监视、控制和支持其日常操作的信息系统。医院使用信息系统来管理患者的电子健康记录，安排预约，并补充补给。呼叫中心正在使用交互式语音应答(IVR)系统为客户提供自动化服务。这些系统还可以收集描述客户在联系系统时经历的每一项活动的事件数据，跟踪资源利用情况，并记录材料和其他物流实体的移动。在我的博士学位中，我创造了术语队列挖掘，这是一组使用事件数据实现自动化数据驱动的业务分析的技术。具体地说，这些数据被用来自动构建底层系统的排队模型，而人工干预最少。到目前为止，队列挖掘主要关注预测分析，即在给定现有系统(没有干预和变化)的情况下，队列挖掘的目标是预测其未来的行为。会议的主要重点是回答这样的问题：“在现有资源的情况下，我的急诊科下周一会有什么表现？”相反，我的目标是开发一种新的队列挖掘方法来回答说明性问题，这将涉及到对希望引入底层系统的更改的分析。拟议研究计划的长期目标是为各种服务系统中的说明性业务分析开发队列挖掘方法，从而以最小的努力生成健壮、客观和可调的模型。

Business analytics, the use of data-driven methodologies to support decisions, plays a central role in organizations. From descriptive analytics that focuses on graphically depicting and statistically detecting problems in the existing system, through predictive analytics that aims at forecasting the system's behavior in the future, to prescriptive analytics that targets system improvement, business analytics became backbone of system and process analysis and optimization. In the proposed research, I shall focus on prescriptive analytics in service systems, such as hospitals, public transportation systems, and datacenters. I will target answering questions such as "how would the Emergency Department behave if we would add 5 nurses during last night's shift?", "when would the bus arrive to its destination had it been dispatched 5 minutes earlier?", and "how can we optimally schedule the datacenter to achieve predefined service level agreements?" Answering these questions in a data-driven fashion is the ultimate goal of evidence-based analysis and optimization in any organization and specifically, in services. In the modern era, most service systems employ information systems that monitor, control, and support their day-to-day operations. Hospitals use information systems to manage electronic health records of patients, schedule appointments, and replenish supplies. Call centers are using interactive voice response (IVR) systems to provide automated services to their customers. These systems can also collect event data that describe every activity that customers experience while contacting the system, track resource utilization, and record movements of materials and other logistic entities. In my PhD, I coined the term Queue mining, which is a set of techniques that enable automated data-driven business analytics using event data. Specifically, the data is used to construct queueing models of the underlying system, automatically, with minimal human involvement. Thus far, queue mining has been mainly focusing on predictive analytics, i.e., given the existing system (without interventions and changes), queue mining would aim at predicting its future behavior. The main emphasis has been on answering questions such as "how would my emergency department behave next Monday, with the existing resources?". Instead, I aim at developing a novel queue mining approach to answer prescriptive questions, which will involve the analysis of changes that one wishes to introduce into the underlying system. The long-term objective of the proposed research program is to develop queue mining methodologies for prescriptive business analytics in a variety of service systems that would produce robust, objective, and tunable models in minimal effort.