Concrete water cement ratio prediction system using random forest regression

This research focused on the development of a concrete water-cement ratio prediction system to maintain a balance in term of the amount of water necessary to achieve optimum concrete strength. In most cases, concrete water cement mixture is usually based on trial and error methods and particularly in the regions where varying environmental conditions affect concrete properties. This method often leads to inconsistent results, resulting in material waste and increased costs. This study employed Random Forest Regression (RFR) model, benchmarked with three other models (Deep Neural Network, Support Vector Regression (SVR), Gradient Boosting Regression (GBR)) for predicting concrete water-cement ratio. The study collected a dataset of seventy-two (72) samples from laboratory experiments investigating the influence of various factors on the strength of normal concrete. This dataset was then augmented to 4000 samples, which were used to develop the predictive model. The collected data was preprocessed and split into 80:20 for training and testing respectively. The result obtained from the proposed RFR model exhibited the best predictive performance among the evaluated ML and DNN algorithms, achieving R2 of 1.00. It outperformed DNN (R2 = 0.84), SVR (R2 = 0.45), and GBR (R2 = 0.99) on the test dataset. The residual was centered around zero and no indication of overfitting was recorded on the training dataset thereby, validating the robustness and effectiveness in prediction. The results obtained indicate that the proposed system is capable of improving the quality, durability and performance of concrete in the construction industry.