Beyond Rational Illusion: Behaviorally Realistic Strategic Classification

Xinpeng Lv, Haotian Wang, Renzhe Xu, Yunxin Mao, Yang Shi, Cyuanlong, Siyang Gao, Xinwang Liu, Wenjing YangSubmission to ICLR 2026 · Rejected

Abstract

Strategic classification (SC) studies the interaction between decision models and agents who strategically manipulate their features for favorable outcomes. Existing SC frameworks typically rely on the idealized assumption that agents are perfectly rational. However, evidence from behavioral economics and psychology consistently shows that real-world decision-making is often shaped by cognitive biases, deviating from pure rationality. Our work challenges the rational-agent paradigm and formally characterizes its limitations. We introduce the behaviorally realistic strategic classification problem, where manipulations arise from realistic psychological and behavioral biases. To address this problem, we extend prospect theory and behavioral economics into the strategic classification settings and propose a Prospect-Guided Strategic Framework (Pro-SF). Specifically, to reflect agents' actual manipulation in practice, our framework incorporates asymmetry between benefits and costs, different subjective reference points, and irrational probability distortion into the Stackelberg game formulation of SC. This paradigm shift redefines both the agent’s strategic manipulation dynamics and the classifier’s optimization objective. Experiments on synthetic and real-world datasets highlight Pro-SF as a behaviorally grounded approach to strategic classification, bridging machine learning and behavioral economics for more reliable deployment in practice.
BibTeX
@inproceedings{lv2026beyond,
  author = {Xinpeng Lv and Haotian Wang and Renzhe Xu and Yunxin Mao and Yang Shi and Cyuanlong and Siyang Gao and Xinwang Liu and Wenjing Yang},
  title = {{Beyond Rational Illusion: Behaviorally Realistic Strategic Classification}},
  booktitle = {International Conference on Machine Learning},
  year = {2026},
  url = {https://openreview.net/forum?id=tm6K5K4t25},
}
Figure