A Probabilistic preference learning approach for multiple criteria ranking in dynamic decision context

We address the challenge of multiple criteria ranking in dynamic decision contexts, where a decision maker’s (DM’s) preferences evolve in response to changing environments. Traditional ranking methods assume static preferences, but real-world scenarios often involve fluctuating decision factors, necessitating a more flexible approach. We propose a novel probabilistic preference learning framework using a linear Gaussian state space model to capture the DM’s evolving preferences. The model tracks time-varying contingent preferences and offers decision recommendations accordingly. We develop an efficient inference algorithm based on machine learning techniques to estimate value-based model parameters. Its effectiveness is demonstrated via practical application in the military context and computational experiments, comparing the novel approach with the state-of-the-art preference learning methods.