Ensemble Regressors for Half Cell Potential Prediction

Abstract

This study addresses the critical issue of steel corrosion in concrete structures, a major concern in the construction industry. By integrating advanced machine learning techniques, particularly ensemble methods, the research aims to enhance the accuracy and reliability of corrosion prediction models for reinforced concrete structures. Through experimentation and meticulous data collection, key input parameters such as distances from the anode, relative humidity, temperature, and concrete age were identified. Various ensemble learning methods including Boosted Trees, Bagged Trees, and Optimizable Ensembles were employed and evaluated using performance metrics such as RMSE, R-squared, MSE, MAE, prediction speed, and training time. LSBoost with Bayesian optimization emerged as the top-performing method, achieving the lowest RMSE of 0.018097, highest R-squared of 0.97, lowest MSE of 0.00032752, and smallest MAE of 0.013769. Despite its longer training time, LSBoost with Bayesian optimization offers superior predictive accuracy compared to other methods, warranting consideration for applications where accuracy is paramount. Bagged Trees and Boosted Trees also demonstrated good performance, balancing prediction speed and accuracy, making them suitable for time-sensitive applications. This research provides valuable insights for developing cost-effective maintenance and rehabilitation strategies, ultimately improving the durability and strength of concrete structures.