Data-Driven Decision-Making ((DM)-M-3): Framework, Methodology, and Directions

A decision problem, according to traditional principles, is approached by finding an optimal solution to an analytical programming decision model, which is known as model-driven decision-making. The fidelity of the model determines the quality and reliability of the decision-making; however, the intrinsic complexity of many real-world decision problems leads to significant model mismatch or infeasibility in deriving a model using the first principle. To overcome the challenges that are present in the big data era, both researchers and practitioners emphasize the importance of making decisions that are hacked up by data related to decision tasks, a process called data-driven decision-making ((DM)-M-3). By building on data science, not only can decision models be predicted in the presence of uncertainty or unknown dynamics, but also inherent rules or knowledge can he extracted from data and directly utilized to generate decision solutions. This position paper systematically discusses the basic concepts and prevailing techniques in data-driven decision-making and clusters-related developments in technique into two main categories: programmable data-driven decision-making (P-(DM)-M-3) and nonprogrammable data-driven decision-making (NP-(DM)-M-3). This paper establishes a (DM)-M-3 technical framework, main methodologies, and approaches for both categories of (DM)-M-3, as well as identifies potential methods and procedures for using data to support decision-making. It also provides examples of how (DM)-M-3 is implemented in practice and identifies five further research directions in the (DM)-M-3 area. We believe that this paper will directly support researchers and professionals in their understanding of the fundamentals of (DM)-M-3 and of the developments in technical methods.