Predictive models influence strategic decisions in wildlife and security settings
Strategic Decision Focused Learning
Computer Science and Game Theory
Summary
Predictive models often help people make better choices, but when these choices involve smart players reacting to the predictions, things can get tricky. The authors examine what happens when predictions affect strategic decisions, like poachers reacting to predictions of animal locations. They find improving prediction accuracy does not always lead to better outcomes because people adapt to the predictions. This work offers new ways to design models that consider how decision-makers might change their strategies in response.
What this means in practice
- •For wildlife protection teams: Design predictive patrol plans that anticipate how poachers change tactics in response to location forecasts.
- •For security operations teams: Improve infrastructure defense by predicting attacker strategies that adapt to revealed threat assessments.
Authors
Tinashe Handina, Yuehan Diao, Adam Wierman, Eric Mazumdar
Abstract
Machine learning (ML) predictions are increasingly being used to guide decision-making, giving rise to the problem of decision-focused learning (DFL) where predictors are optimized for downstream decision quality rather than accuracy alone. However, most existing work assumes a single decision-maker optimizing in isolation. This paper formalizes strategic decision-focused learning, where an ML system predicts an exogenous state that some agents observe before playing a game. For example, a park ranger may predict wildlife locations to allocate anti-poaching patrols against strategic poachers. While the exogenous state is unaffected by agent actions, predictions influence agents' strategies and the resulting equilibrium. We find that strategic considerations fundamentally change the learning problem. In particular, we show the prediction accuracy-equilibrium payoff landscape can be non-monotonic, i.e., better predictions can degrade performance. We propose algorithmic approaches to address these challenges and validate them across benchmarks in wildlife conservation and infrastructure protection. Our theory and experiments highlight the importance of accounting for strategic interactions when designing predictors.