Personalized Decision Support Driven by Similarity Metrics

决策是包括医学、公共卫生、政治、经济、商业、零售和体育在内的各个领域的重要方面。虽然最佳的决策植根于定量证据是可取的，挑战出现在缺乏既定的决策指导方针支持的科学研究。如果没有明确的指导方针，决策者就会求助于以前的培训、当地文化和轶事经验，这往往会导致有偏见的推理。由于信息技术的进步，今天电子数据的大量生产和储存为一种有吸引力的替代办法打开了大门，即数据驱动的决策，它从有价值的电子数据中提取相关知识，并立即将其提供给决策者。 我建议研究的数据驱动决策的一个有前途的途径是基于电子数据中类似案例分析的个性化决策支持。设想的核心技术是计算相似性度量（SM），客观地量化两个案例彼此相似的程度。根据应用程序，“案例”可以是患者、组织、消费者或社区。基于SM，个性化的决策支持范例首先确定过去的情况下，就像一个正在调查的当前情况下。通过分析这些相似案例在其等效时间点之后发生的情况，可以为个性化决策提供信息。具体而言，SM可以帮助预测当前案例的未来，或者通过调查过去决策的影响来促进决策分析。这项拟议的研究假设纵向模式（即，数据作为时间的函数），除了静态信息之外，在SM中起着至关重要的作用，因为它们捕获动态特性。由于互联网和无处不在的移动的设备，我们现在拥有了强大的计算机、廉价的数据存储和新颖的数据源，相似性的概念可以被收紧，从传统的一刀切的决策转向量身定制的实时决策。毕竟，每个案例在某种程度上都是独特的。 这项研究的主要成果将是一个完整的决策支持基础设施，包括SM，数据仓库，计算节点和软件决策支持工具，有效的用户界面和数据可视化。将利用公共数据库，以避免耗时的数据收集工作。数学和统计建模以及人工智能将被用来开发通用和特定问题的SM，而高性能计算将成为决策支持工具的支柱。SM与个性化决策支持的无缝集成以及与决策相关的知识的实时培养是完全新颖的，并且将推动当前最先进的决策支持流程的发展。 这项拟议的研究计划是数据科学的核心，预计在可预见的未来，数据科学将在学术界和工业界创造重要的就业机会。然而，大多数大学目前没有开设数据科学学位课程，因为它是一个相对较新的领域，是计算机科学、统计学和数学等几个不同领域的交叉点。通过让本科生和研究生领导研究的各个方面，拟议研究的重中之重是培养高素质的人才，作为精通数据科学家，具备实际计算技能和理论基础。

Decision making is an important aspect of various fields including medicine, public health, politics, economics, business, retail, and sports. Although optimal decisions rooted in quantitative evidence are desired, challenges arise in the absence of established decision making guidelines supported by scientific research. When there is no clear guideline, decision makers resort to previous training, local culture, and anecdotal experiences, which frequently lead to biased reasoning. Today’s massive production and storage of electronic data thanks to advances in information technology open doors to an attractive alternative, namely data-driven decision making, which extracts relevant knowledge from valuable electronic data and instantaneously delivers it to decision makers. One promising avenue in data-driven decision making that I propose to investigate is personalized decision support based on analysis of similar cases in electronic data. The envisioned core technology is computation of similarity metrics (SMs) that objectively quantify the extent to which two cases are similar to each other. Depending on the application, a ‘case’ can be a patient, organization, consumer, or community. Based on an SM, the personalized decision support paradigm first identifies past cases that are just like a current case under investigation. By analyzing what happened to those similar cases after their equivalent point in time, it is feasible to inform personalized decision making. Specifically, SMs can help forecast the future of the present case or facilitate decision analysis via investigating the effects of past decisions. This proposed research hypothesizes that longitudinal patterns (i.e., data as a function of time), in addition to just static information, play a crucial role in SMs since they capture dynamic characteristics. Now that we have powerful computers, inexpensive data storage, and novel data sources thanks to the Internet and ubiquitous mobile devices, the concept of similarity can be tightened to move away from traditional one-size-fits-all decisions to tailor-made real-time decisions. After all, every case is unique at some level of detail. The primary outcome of this proposed research will be a complete decision support infrastructure encompassing SMs, data warehouses, computational nodes, and software decision support tools with effective user interfaces and data visualization. Public databases will be utilized in order to bypass time-consuming data collection. Mathematical and statistical modeling as well as artificial intelligence will be leveraged to develop both generic and problem-specific SMs, while high-performance computing will be the backbone of the decision support tools. The seamless integration of SMs with personalized decision support and real-time cultivation of knowledge relevant to decision making are completely novel and will push the envelope of the current state-of-the-art decision support processes. This proposed research program is at the core of data science, which is expected to be a significant job creator in the foreseeable future in both academia and industry. However, most universities currently do not operate degree programs in data science because it is a relatively new field in the intersection of several distinct fields including computer science, statistics, and mathematics. By having undergraduate and graduate students lead all aspects of the research, a top priority in the proposed research is to train highly qualified personnel as proficient data scientists equipped with both hands-on computing skills and theoretical foundations.