ABSTRACT
Cracking is one of the main problems in concrete structures and is affected by various parameters. The step-by-step laboratory method, which includes casting specimens, curing for a certain period, and testing, remains a source of worry in terms of cost and time. Novel machine learning methods for anticipating the behavior of raw materials on the ultimate output of concrete are being introduced to address the difficulties outlined above such as the excessive consumption of time and money. This work estimates the splitting-tensile strength of concrete containing recycled coarse aggregate (RCA) using artificial intelligence methods considering nine input parameters and 154 mixes. One individual machine learning algorithm (support vector machine) and three ensembled machine learning algorithms (AdaBoost, Bagging, and random forest) are considered. Additionally, a post hoc model-agnostic method named SHapley Additive exPlanations (SHAP) was performed to study the influence of raw ingredients on the splitting-tensile strength. The model's performance was assessed using the coefficient of determination (R2), root mean square error (RMSE), and mean absolute error (MAE). Then, the model's performance was validated using k-fold cross-validation. The random forest model, with an R2 of 0.96, outperformed the AdaBoost models. The random forest models with greater R2 and lower error (RMSE = 0.49) had superior performance. It was revealed from the SHAP analysis that the cement content had the highest positive influence on the splitting-tensile strength of the recycled aggregate concrete and the primary contact of cement is with water. The feature interaction plot shows that high water content has a negative impact on the recycled aggregate concrete (RAC) splitting-tensile strength, but the increased cement content had a beneficial effect.
ABSTRACT
Researchers and engineers are presently focusing on efficient waste material utilization in the construction sector to reduce waste. Waste marble dust has been added to concrete to minimize pollution and landfills problems. Therefore, marble dust was utilized in concrete, and its prediction was made via an artificial intelligence approach to give an easier way to scholars for sustainable construction. Various blends of concrete having 40 mixes were made as partial substitutes for waste marble dust. The ultrasonic pulse velocity of waste marble dust concrete (WMDC) was compared to a control mix without marble dust. Additionally, this research used standalone (multiple-layer perceptron neural network) and supervised machine learning methods (Bagging, AdaBoost, and Random Forest) to predict the ultrasonic pulse velocity of waste marble dust concrete. The models' performances were assessed using R2, RMSE, and MAE. Then, the models' performances were validated using k-fold cross-validation. Furthermore, the effect of raw ingredients and their interactions using SHAP analysis was evaluated. The Random Forest model, with an R2 of 0.98, outperforms the MLPNN, Bagging, and AdaBoost models. Compared to all the other models (individual and ensemble), the Random Forest model with greater R2 and lower error (RMSE, MAE) has a superior performance. SHAP analysis revealed that marble dust content has a positive and direct influence on and relationship to the ultrasonic pulse velocity of concrete. Using machine learning to forecast concrete properties saves time, resources, and effort for scholars in the engineering sector.
ABSTRACT
This work evaluates the performance of three machine learning (ML) techniques, namely logistic regression (LGR), linear regression (LR), and support vector machines (SVM), and two multi-criteria decision-making (MCDM) techniques, namely analytical hierarchy process (AHP) and the technique for order of preference by similarity to ideal solution (TOPSIS), for mapping landslide susceptibility in the Chitral district, northern Pakistan. Moreover, we create landslide inventory maps from LANDSAT-8 satellite images through the change vector analysis (CVA) change detection method. The change detection yields more than 500 landslide spots. After some manual post-processing correction, the landslide inventory spots are randomly split into two sets with a 70/30 ratio for training and validating the performance of the ML techniques. Sixteen topographical, hydrological, and geological landslide-related factors of the study area are prepared as GIS layers. They are used to produce landslide susceptibility maps (LSMs) with weighted overlay techniques using different weights of landslide-related factors. The accuracy assessment shows that the ML techniques outperform the MCDM methods, while SVM yields the highest accuracy of 88% for the resulting LSM.
