AI-Driven predicting and optimizing lignocellulosic sisal fiber-reinforced lightweight foamed concrete: A machine learning and metaheuristic approach for sustainable construction

• Improved sisal fiber reinforced lightweight foamed concrete by machine learning and metaheuristic algorithms. • Developed a deep neural network (DNN) enhanced by the Grey Wolf Algorithm (GWO) and the Slime Mould Algorithm (SMA) for the prediction of tensile strength. • Assessed six prediction models, with DNN-IGWO surpassing conventional machine learning methods in both accuracy and generalization. • SMA converged efficiently in 216 iterations, attaining an ideal tensile strength of 4.16 MPa, exceeding typical experimental design forecasts of 3.8 MPa. • The low standard deviation (SD = 2.15 × 10⁻⁵) confirms the stability and resilience of the optimized mix design. This research investigates the application of machine learning (ML) and metaheuristic optimization to improve the mechanical properties of sisal fiber-reinforced foamed concrete. A Deep Neural Network (DNN) was developed, optimized using the Grey Wolf Optimizer (GWO) and the Slime Mould Algorithm (SMA), to predict and optimize material properties. Thirty-four data points obtained through experimentation were utilized to train, test, and validate various machine learning models for the prediction of tensile strength (TS). Six predictive models were assessed for accuracy and generalization: Support Vector Machine (SVM), Decision Tree (DT), K-Nearest Neighbor (KNN), Linear Model (LM), Dragonfly Algorithm-based Deep Neural Network (DNN-DA), and Improved Grey Wolf Optimizer-based Deep Neural Network (DNN-IGWO). DNN-IGWO demonstrated enhanced predictive performance, significantly exceeding that of traditional machine learning models. The optimization process directed by the metaheuristic approach utilizing SMA demonstrated swift convergence in 216 iterations, identifying the optimal mix proportions of cement, water-to-cement ratio, sand, and sisal fiber content. The optimized composition achieved a tensile strength of 4.16 MPa, representing a 9.5 % enhancement compared to conventional experimental methods, which yielded 3.8 MPa. Statistical validation demonstrated the model's stability and reliability, evidenced by a notably low standard deviation (SD = 2.15 × 10⁻⁵), indicating minimal variability in predictions. This study illustrates a comprehensive AI-based framework for the optimization of cementitious materials, effectively integrating experimental and computational methodologies. The proposed approach minimizes dependence on labor-intensive trial-and-error testing and enhances the sustainability of construction materials by utilizing sisal fibers. The results emphasize the capabilities of metaheuristic-enhanced deep learning models in optimizing high-performance, environmentally friendly concrete mix design, thereby facilitating future advancements in intelligent material formulations for sustainable infrastructure.