Fine Tuning the Prediction of the Compressive Strength of Concrete : A Bayesian Optimization Based Approach

This study demonstrates the efficacy of Bayesian Optimization to improve the performance of machine learning models for predicting the strength properties of concrete specimens. There exists several machine learning models for predicting concrete compressive strength. Most of these models have hyper-parameters that, if properly tuned, will improve the models even further. Although the selection of these hyperparameters has a significant impact on the performance of models, the process of hyperparameter tuning is often ignored in previous works. In the first phase of the study, five machine learning models(SVR, ABR, RFR, GBR and KNN Regressor) were compared on the basis of rmse, mae and r-squared value on the test set. Two best performing models(GBR and RFR) were selected among the five models for further improvement. In the second phase, hyperparameter tuning by Bayesian Optimization method was done on these two models. Experimental results testify that Bayesian Optimization on these two models improved their prediction performance further.