Conventional and Ensemble Machine Learning Techniques to Predict the Compressive Strength of Sustainable Concrete

Oil palm shells (OPS) can be utilized as a sustainable substitute for natural coarse aggregates in the making of concrete. Due to its various advantages in concrete manufacturing, including environmental sustainability, lower density, good insulating qualities, and lower cost. The use of appropriate additives, as well as proper design and mix proportions, can help to optimize the mechanical characteristics of concrete containing OPS. To optimize mix design, anticipate mechanical characteristics either an exhaustive set of experiments or soft computing techniques are required. To that objective, various soft computing techniques were used in this study. Firstly, a correlation matrix between various features of sustainable concrete was established. Machine learning (ML) models were developed for predicting the compressive strength (CS) of OPS-based concrete composite. Various ML models such as decision tree (DT) was developed as a conventional machine learning (CML) model, whereas Random Forest (RF), AdaBoost (AdB), and Gradient Boosting (GB) were developed as ensemble machine learning (EML) models. Hyperparameter tuning was also performed to enhance each model’s performance. As a result, all developed models predicted the CS of concrete containing OPS effectively. Models were examined using performance evaluation methods, and it was found that the GB model fared the best in both training and testing phases, with the lowest RMSE and MAE of 0.428 and 0.341, respectively, with higher R² as 0.998. Simultaneously, the RF's predicted performance for this data was determined to be inferior, with RMSE and MAE of 2.096 and 1.578, respectively, and lower R² value as 0.953.