Prediction of concrete’s compressive strength using machine learning algorithms

Abstract

The study focuses on predicting the behavior of Lightweight Aggregate Geopolymer Concrete (LWAGC) using machine learning approaches. The performance measures Mean Squared Error (MSE) and Mean Absolute Error (MAE) were used to assess many machine learning models. The XGBoost algorithm outperformed other models, with the lowest MSE of 17.34 and the lowest MAE of 2.87. With an MSE of 34.02 and an MAE of 4.46, the LSTM-ANN Hybrid model likewise fared well. The LSTM-CNN Hybrid model, on the other hand, had a higher MSE of 293.92 and a MAE of 13.82, showing space for improvement. When compared to the top-performing algorithms, linear regression, ANN, and CNN models had greater MSE and MAE. These findings emphasize the accuracy with which machine learning approaches, notably XGBoost and LSTM-ANN Hybrid, anticipate LWAGC behavior. Precise prediction allows for the construction of lightweight concrete buildings, which have advantages such as lower density, higher fire resistance, and a lower carbon footprint. This study adds to the body of knowledge on LWAGC by proving the efficiency of machine learning in predicting the behavior of lightweight concrete. The findings may be used to improve the design of lightweight concrete buildings, encouraging sustainable and resilient construction methods in civil engineering. The training and testing data used in this study were divided into a 70% training set and a 30% testing set.