Predictive analysis of concrete slump using a stochastic search-consolidated neural network

Area of Science:

• Civil Engineering
• Computational Intelligence
• Materials Science

Background:

• Concrete slump is a critical indicator of workability, but traditional measurement methods are costly and complex.
• Metaheuristic-based machine learning offers efficient indirect models for approximating concrete slump.
• Evaluating advanced optimization algorithms is essential for updating and improving concrete slump prediction models.

Purpose of the Study:

• To enhance a multi-layer perceptron neural network (NN-MLP) using the Stochastic Fractal Search (SFS) optimization algorithm.
• To develop a cost-efficient and accurate model for predicting concrete slump based on mixture ingredients and curing age.
• To compare the performance of the SFS-enhanced model against other metaheuristic algorithms like EHO and SMA.

Main Methods:

• Utilizing a multi-layer perceptron neural network (NN-MLP) architecture.
• Employing the Stochastic Fractal Search (SFS) algorithm to optimize NN-MLP parameters.
• Validating the SFS-NN-MLP model's predictive accuracy against Elephant Herding Optimization (EHO) and Slime Mould Algorithm (SMA) enhanced NN-MLP models.

Main Results:

• The SFS-NN-MLP model demonstrated superior accuracy in predicting concrete slump compared to EHO-NN-MLP and SMA-NN-MLP.
• Quantitative metrics showed the SFS-NN-MLP achieved the lowest mean square error (5.6526) and mean absolute error (4.6657).
• The SFS-NN-MLP model achieved a higher percentage-Pearson correlation coefficient (78.06%) indicating a stronger predictive relationship.

Conclusions:

• The SFS-NN-MLP model provides a highly accurate and reliable method for approximating concrete slump.
• Stochastic Fractal Search is an effective optimizer for enhancing neural network performance in concrete technology applications.
• The proposed SFS-NN-MLP model is recommended for practical, cost-efficient concrete slump estimation in construction projects.