Machine learning methods for assessing value-of-information

Abstract

One of the most useful features of decision analysis is its ability to distinguish between constructive and wasteful information gathering. Value-of-information (VOI) and sequential information gathering (Value-of-Flexibility, VOF) analysis evaluates the benefits of collecting additional information before making a decision. Traditionally, VOI has been assessed by constructing a decision tree or influence diagram model where a Bayesian framework has been used to update probabilities given new information. In this research, we evaluate the use of machine learning (ML) methods such as Ordinary-Least-Square (OLS), Random Forest (RF), K-Nearest Neighbor (KNN), Support Vector Regression (SVR), and Extreme Gradient Boost (XGB) for VOI calculations. In this study, VOI will be estimated using a simulation-regression approach. In the simulation-regression approach, VOI is computed by simulating the model parameters, the data and prospect values, then regressing the prospect values on the data (Eidsvik, Mukerji et al. 2015, Eidsvik, Dutta et al. 2017, Dutta, Mukerji et al. 2019). Simulation-regression approach is considered to be one solution to overcome the computational issue by constructing efficient approximations for the VOI. In addition, VOI and Value-of-Flexibility (VOF) is implemented on case study of estimating CO2 storage capacity of Utsira formation located in North Sea using simulation-regression approach.