Optimizing the compressive strength prediction of geopolymer lime mortars using the PCA-ELM artificial intelligence model

Abstract

This study proposes the use of geopolymer lime mortar, activated with NaOH and Na2SiO3 alkalis, and made from lime, fly ash, brick aggregate, and blast furnace slag (BFS), as an alternative to Portland cement-based concrete. The geopolymer lime mortar samples used in the experimental analysis were produced under controlled laboratory conditions. Compressive strength tests were conducted on the produced samples. The sample with the highest BFS content yielded the best compressive strength results. However, experimental studies are time-intensive. To shorten the experimental time and minimize the material and equipment costs associated with the experiments, a hybrid regression algorithm was proposed for the prediction of compressive strength. Instead of labratory tests the compressive strength of the produced samples was determined using a hybrid regression algorithm has never been used before for this purpose in the literature. This hybrid algorithm is the principal components analysis extreme learning machine algorithm obtained by integrating the PCA method, an effective feature selection method in machine learning, and the ELM method, a regression method that has increased its popularity in recent years. The performance of the proposed algorithm has been compared with other neural network models such as Artificial Neural Network and ELM algorithms and also compared with frequently used algorithms such as random forest regressor, ada boosting, gradient boosting, and extreme gradient boosting algorithms. The results obtained demonstrated the ability of the proposed PCA-ELM algorithm to capture complex relationships within the data by exhibiting superior performance compared to commonly used methods in compressive strength estimation of geopolymer lime mortar.