Search

In designing risk assessment algorithms, many scholars promote a "kitchen sink" approach, reasoning that more information yields more accurate predictions. We show, however, that this rationale often fails when algorithms are trained to predict a proxy of the true outcome, for instance, predicting arrest as a proxy for criminal behavior. With this "label bias," one should exclude a feature if its correlation with the proxy and its correlation with the true outcome have opposite signs, conditional on the other model features. This criterion is often satisfied when a feature is weakly correlated with the true outcome, and, additionally, that feature and the true outcome are both direct causes of the proxy outcome. For example, criminal behavior and geography may be weakly correlated and, due to patterns of police deployment, direct causes of one's arrest record-suggesting that excluding geography in criminal risk assessment will weaken an algorithm's performance in predicting arrest but will improve its capacity to predict actual crime.

Current mitigation strategies to offset marine plastic pollution, a global concern, typically rely on preventing floating debris from reaching coastal ecosystems. Specifically, clean-up technologies are designed to collect plastics by removing debris from the aquatic environment such as rivers and estuaries. However, to date, there is little published data on their potential impact on riverine and estuarine organisms and ecosystems. Multiple parameters might play a role in the chances of biota and organic debris being unintentionally caught within a mechanical clean-up system, but their exact contribution to a potential impact is unknown. Here, we identified four clusters of parameters that can potentially determine the bycatch: (i) the environmental conditions in which the clean-up system is deployed, (ii) the traits of the biota the system interacts with, (iii) the traits of plastic items present in the system, and, (iv) the design and operation of the clean-up mechanism itself. To efficiently quantify and assess the influence of each of the clusters on bycatch, we suggest the use of transparent and objective tools. In particular, we discuss the use of Bayesian Belief Networks (BBNs) as a promising probabilistic modelling method for an evidence-based trade-off between removal efficiency and bycatch. We argue that BBN probabilistic models are a valuable tool to assist stakeholders, prior to the deployment of any clean-up technology, in selecting the best-suited mechanism to collect floating plastic debris while managing potential adverse effects on the ecosystem., (Copyright © 2021 The Authors. Published by Elsevier Ltd.. All rights reserved.)