A value of prediction model to estimate optimal response time to threats for accident prevention

Abstract

This paper presents a novel value of (imperfect) prediction (VoP) model to estimate optimal response time to a threat that may result in an accident. The proposed VoP model is based on information value theory and considers both prediction accuracy and action failure probability over time. The optimal response time is dependent on parameters: the ratio between the accident cost and response action cost, accident probability, action failure probability, prediction performance, and response strategy (a series of sequential responses or a single response). A case study of iceberg management is presented to demonstrate the proposed approach; a sensitivity study is done to evaluate how optimal response time changes with those parameters. The case study show that it is reasonable to respond as early as possible if the threat can lead to a serious accident, while the response can be postponed when the potential consequence is moderate. In addition, the proposed VoP model is proven able to calculate accuracy requirements, thresholds for tolerating risk and acting precautionarily, and maximum investment in accident prevention. Imperfect prediction can lower risk acceptance threshold and higher the threshold of being precautionary; and it is reasonable to increase action cost.