PSI - Issue 70

Sahil Sehrawat et al. / Procedia Structural Integrity 70 (2025) 394–400

396

to forecast the compressive strength (CS) of concrete. The research demonstrated that machine learning is capable of forecasting various mechanical attributes. Nguyen et al. (2021) employed SML methods to forecast the tensile and compressive strengths of high-performance concrete. It was determined that ensemble SML demonstrated greater accuracy compared to individual methods. Machine learning enhances models that struggle, such as decision trees and multilayer perceptron’s . As a result, numerous researchers released SML techniques that enhance the accuracy of material property estimations. This matter necessitates comprehensive investigation. 2. Literature Review 2.1. FindingandAnalyzingData SML algorithms require several input variables to succeed (Sufian et al., 2021). Table 1 of Supplementary Materials displays GPC article projected CS. GPC CS papers were tallied using comparable services. Many studies examined GPC features, but this one utilised CS-data model. Fly ash, GGBS, NaOH, fine aggregate, gravel types (10/20 mm), NaOH molarity, and water/solids ratio were model inputs. Each model gives CS alone. Many data sets and inputs impact model output (Ahmad et al., 2021d). The research employed SML on 371 data points. Every input factor is included in Table 1. All variable relative occurrence dispersion is shown in Figure 2. This counts all observations for each value or collection. Probability distributions are fundamental to statistics. Table 1. Explanatory statistics of select model parameters. FlyAsh GGBS NaOH Water Gravel 10/20 FineAgg. (kg/m 3 ) (Kg/m 3 ) Molarity Ratio Mm (kg/m 3 ) (kg/m 3 ) Mean 174.54 225.25 8.15 0.35 737.47 730.89 Mode 0 0 10.1 0.52 0 652 Median 123 310 9.21 0.35 799 738 Standard Deviation 168.05 162.31 4.57 0.12 358.65 131.07 Sum 63,290.03 80,928.05 2000.11 125.78 267,665.03 265,047.79 Range 523 450 19.00 0.63 1298 901 Maximum 533 449 20.00 0.64 1301 1359 Minimum 0 0 1.00 0 0 459 Standard Error 8.79 8.58 0.25 0.02 18.79 6.97 2.2. Analysis of Techniques Employed This study used Python, Anaconda Navigator, and built DT and RF SML algorithms. DT and RF models ran on Spyder version 4.3.5. These algorithms anticipate outcomes from inputs. The projected material strength, durability, and temperature matter (Ahmad et al., 2021e; Song et al., 2021b). One output and nine inputs were simulated. All models had valid and accurate R² values for expected outcomes. The model measures response variable variance with R², the coefficient of determination. Data-model alignment is assessed, whereas 1 indicates a near perfect model-data match (Ahmad et al., 2021c). SML methods from this study are below. All models were checked using statistics, KFCV, error inquiry, RMSE, and MAE. Input parameters impacted projected conclusions in sensitivity analysis. Fig. 2 displays research flowchart.