Application of machine learning for the prediction of atmospheric corrosion in the metropolitan area of mexico city

Abstract

In this work, Machine Learning (ML) is applied to predict atmospheric corrosion in the Metropolitan Zone of Mexico City. For this purpose, mass loss is measured as a dependent variable associated with the independent variables relative humidity, wetting time, temperature and sulfur dioxide deposition time in 12 stations of the study site and with the generated database. 

ML models are used with some supervised learning tools, such as: Neural Networks (NN), Regression Trees (RT), Optimized Regression Tree (ORT), Regression Ensemble (RE), Support Vector Machine (SVM) and Linear Regression (LR). For this problem, Neural Networks (NN) have the best results, with a Correlation Coefficient R² = 0.9814 and a Mean Square Error MSE = 37.9.

 The main results allow us to determine that the proposed framework can be extended to predict the behavior of other complex problems.