FB-XGBoost: A Hybrid Model for Accurate Concrete Compressive Strength Prediction

Models with high precision and strong generalisation capabilities can better solve prediction problems in daily life. However it is a challenge to select and build a highly adaptable model. In this research, a novel hybrid model (FB-XGBoost model) for concrete compressive strength prediction is proposed via the combination of domain-specific feature engineering and Optimised XGBoost. Feature engineering is used to construct the model that includes the key factors affecting concrete strength and Bayesian optimization is applied for hyperparameter tuning of the model to maximize the performance of the model. The experimental results on a real world dataset show that the proposed model surpasses the traditional method like Ridge Regression, Linear Regression and also other machine learning or deep learning models like SVR, Random Forests. The model has MSE equal to 16.3185, MAE equal to 2.8507, and R2 equal to 0.9418, which shows very high accuracy with respect to predicting the concrete compressive strength. This hybrid model supplies a robust tool for precise concrete strength prediction and it can be extended to more predictive tasks.