T. Ekin, W.H. Caballero, R. Naveiro, D. Rios Insua
Texas State University, Texas, United States
Poster stand number: T113
Keywords: Decision Analysis, Adversarial Forecasting, Adversarial Risk Analysis ,Batch Data, Hidden Markov ModelsForecasting methods typically assume clean and legitimate data streams. However, adversaries may attempt to influence data and alter forecasts, which in turn may impact decisions. This poster presents a decision theoretic approach for adversarial forecasting. Proposed adversarial risk analysis based framework allows incomplete information and adversarial perturbations on the forecasting output. We solve the adversary’s poisoning decision problem where he manipulates batch data inputted into forecasting methods. In particular, adversarial auto-regressive and hidden Markov models are presented in detail and demonstrated with examples using real world data. The findings show the vulnerability of forecasting models under adversarial activity. We discuss potential defender strategies to improve the security of existing decision frameworks that use forecasting method outputs.