Sparse Probability Assessment Heuristic Based on Orthogonal Matching Pursuit

Abstract

Probability assessment via expert elicitation or statistical analysis is a critical step in the decision-analysis process. In many actual applications, the number of uncertainties and the corresponding number of assessments can be quite large. In these cases, the analyst may seek guidance in focusing the assessment on the most important uncertainties. In this paper, we develop a novel heuristic that we call the sparse probability assessment heuristic (SPAH). SPAH, which is based on a well-known method in machine learning known as orthogonal matching pursuit, seeks to identify the preferred alternative while conducting the fewest number of assessments. We test SPAH under a variety of conditions and compare its performance to standard practice. In so doing, we show that SPAH is able to identify the optimal alternative while requiring substantially fewer assessments than standard practice.