Data-Driven Approach with Machine Learning to Reduce Subjectivity in Multi-Attribute Decision Making Methods

Abstract

Multi-Attribute Decision Making (MADM) methods are an integral component of trade-off studies which are frequently employed in Systems Engineering when multiple interdependent decision criteria are involved. In MADM methods, each decision criterion is assigned a weight based on how important it is to the Decision-Makers (DMs), and a decision matrix is populated with values representing assessments of each alternative with respect to the decision criteria. MADM methods, therefore, are susceptible to subjectivity due to inherent bias in DM’s preferences where slight fluctuation in stated DM’s preference can drastically impact the outcome. In this paper, we propose a data-driven methodology with Machine Learning to improve the effectiveness of MADM methods by reducing DMs’ subjective biases resulting from criteria weights. In addition, the proposed methodology leverages Exploratory Data Analysis to better determine the type of criteria as cost or benefit, depending upon whether it positively or negatively affects the MADM outcome. A sample trade study example of selecting a metropolitan area based on housing affordability is provided to illustrate how the proposed method is applied to generate data-based true criteria weights and types.