A Comparative Study of AdaBoost and K-Nearest Neighbor Regressors for the Prediction of Compressive Strength of Ultra-High Performance Concrete

One of the most promising materials for concrete buildings is ultra-high-performance concrete (UHPC). Traditional UHPC compositions include significant amounts of cement, silica fume, superplasticizer, and other pricey and carbon-intensive ingredients. In order to develop a more cost-effective and environmentally friendly UHPC using alternative UHPC dosages that utilise locally accessible resources, it is necessary to study the relationships between the UHPC dosage and its resulting properties. This study employs two novel machine learning algorithms, the AdaBoost regressor and K-nearest neighbor, to illustrate the non-linear relationships between dose mixture design and the compressive strength of UHPC. A dataset comprising 810 UHPC mixture collections with 15 input variables, namely cement, slag, fly ash, silica fume, quartz powder, limestone powder, nano silica, water, coarse aggregate, fine aggregate, fibre, superplasticizer, relative humidity, temperature, and age has been used to train the models. After adjusting the regression model, the prediction performance of the two models is comprehensively compared using different performance parameters. The proposed AdaBoost regressor model achieved the most precise prediction during the testing phase, outperforming the K-nearest neighbor regressor, as evident from the statistical results and Taylor diagram. Shapley additive explanations (SHAP) measure feature significance and variable influence on a prediction. The SHAP interpretations matched the typical compressive behaviour of concrete, confirming the typical relationship between machine learning predictions and actual events. The proposed AdaBoost model can be used as a high-performance tool to estimate the compressive strength of ultra-high performance concrete during the design and construction phases of civil engineering projects based on the experimental results.