Hybrid metaheuristic optimized Catboost models for construction cost estimation of concrete solid slabs

Area of Science:

• Construction Management
• Artificial Intelligence in Engineering
• Computational Optimization

Background:

• Accurate construction cost prediction is vital for project success in the competitive industry.
• Traditional methods often struggle with the complexity of cost factors.
• Optimizing machine learning hyperparameters is crucial for predictive accuracy.

Purpose of the Study:

• To develop and evaluate advanced hybrid models for concrete solid slab cost prediction.
• To optimize the performance of the Categorical Boosting (CatBoost) model using metaheuristic algorithms.
• To identify key cost-influencing variables through explainable AI techniques.

Main Methods:

• Hybridization of CatBoost with Phasor Particle Swarm Optimization (PPSO), Dwarf Mongoose Optimization (DMO), and Atom Search Optimization (ASO).
• Hyperparameter optimization (depth, learning rate, iterations) for CatBoost models.
• Performance evaluation using residual error cumulative distribution (REC) curves, scatter plots, violin plots, and quantitative metrics (R², RMSE).
• SHapley Additive exPlanations (SHAP) for variable importance analysis.

Main Results:

• Hybrid models significantly outperformed the standalone CatBoost model.
• ASO-CatBoost achieved the highest accuracy with R² of 0.981 and RMSE of 1.222 $/m².
• SHAP analysis identified Tributary Area and Concrete as the most influential cost predictors.
• The ASO-CatBoost model demonstrated excellent generalization capabilities.

Conclusions:

• Optimized hybrid metaheuristic models provide superior construction cost prediction accuracy.
• ASO-CatBoost is a highly effective tool for predicting concrete solid slab costs.
• A user-friendly Python GUI facilitates practical application of the developed cost estimation model.