Probabilistic Forecast of Concrete Compressive Strength Using ML

Concrete compressive strength is one of the critical indices of concrete characteristics. Researchers have developed several models to predict the concrete compressive strength using various machine learning techniques, which are considered deterministic methods. This paper aims to forecast the concrete compressive strength using a probabilistic prediction method. For this purpose, two boosting algorithms were selected mainly: LightGBM and XGBoost, where their combination with quantile regression has the ability to predict intervals and quantile forecasts. Quantile XGboost outperformed quantile lightGBM in all metrics with a lower quantile score, indicating a better quantile prediction and lower error values for both the RMSE and MAE for all the quantiles except the 0.05 quantile, where quantile lightGBM performed better. In the prediction of 95% intervals, XGboost achieved better performance with a narrow average width equal to 0.4375 and higher coverage of data with a value of the prediction interval coverage probability equal to 0.9369.