A model for predicting corrosion losses of carbon steel for the first year of exposure based on the random forest algorithm

Based on the random forest (RF) algorithm, two models have been obtained for predicting first-year corrosion losses (K 1 ) of carbon steel in an open atmosphere in various regions of the world. The first RF_general model was obtained using the combined databases of the international ISO CORRAG, MICAT, ECE/UN programs and tests in Russia and is designed to assess K 1 in various types of atmosphere in different regions of the world. The second RF_cont model allows you to predict K 1 in the continental regions of the world. The accuracy of K 1 predictions based on RF models and two dose-response functions was compared: the one presented in ISO 9223 standard and the new version developed by IPCE RAS for continental regions. It is shown that the reliability of both RF models is significantly better than the dose-response functions, with the exception of predictions of corrosion losses of steel in Russia with a cold climate.