PSI - Issue 70

Indu Sharma et al. / Procedia Structural Integrity 70 (2025) 380–385

382

combination. This research utilized literature data to conduct multiple trials across different zones, sets, and geometrical configurations. The models only used for input variables and outcomes, so variations in testing settings, layouts, or sample shape did not affect the conclusions of the study. Table 1 presents the descriptive statistics for each input variable. The selected data was also normalized. Normalization of databases. Constructing tables and connecting them through data protection principles reduces duplication and inconsistent dependencies, enhancing the versatility of the database. Concise, objective descriptive statistics provide results for a population or subgroup. Maximum, minimum, and standard deviation demonstrate variance; mean, median, and mode uncover underlying patterns. Table 1 includes all mathematical terms for each model input variable. illustrates the distributions of CS input factors. Displaying input-output parameter correlations and frequency distribution in a diagonal format. The line graph illustrates the connections of positive and negative y-axis parameters through its upward trend corresponding to each x-axis input/output parameter 2.2. Machine Learning Algorithms Employed SVM and ensemble ML algorithms were utilized in Python using Anaconda Navigator to achieve the research objectives. Spyder (4.3.5) executed SVMs. These methods utilize input parameters to forecast outcomes. These methods forecast temperature, strength, and durability song et al. (2015). The model utilized nine inputs and produced one output (CS). The R2 of the expected result indicates the performance of the model. An R2 value close to one signifies an almost perfect alignment between the model and experimental results, while a value approaching zero suggests greater divergence. This study's ML approaches are covered by the following sub-segments. All models were evaluated using RMSE and MAE for k-fold, statistical, and error analysis. Shapley Additive explanations (SHAP) served as a model-independent post hoc technique to evaluate the influence of input factors on GPC CS. Fig. 1 illustrates the research methodology (flowchart).

Fig. 1. Flow-chart of Research Methodology.

Table 1. Input-output descriptive measures. Fine GGBS

FlyAsh

NaOH Water

Gravel Size:

CS

(kg/m3 ) 729.88

(kg/m3)

(kg/m3 )

Molarity Ratio 10/20 mm (kg/m3)

(MPa)

Mean Mode

225.15

174.34

8.14

0.34 0.53 0.34 0.63

737.37

43.28 56.00 42.10 86.08

651 728

0

0

10

0

Median

300 450

120 523

9.2

789

1360

Maximum Minimum Standard

20

1298

459

0

0

1

0

0

8.00

130.97

162.27

167.95

4.56

0.11

358.55

17.87

26,4947.79

Sum

81,728.05 63,286.04 2955.11 124.78 267,664.93

15,710.40

901 6.87

Range

450 8.52

523 8.82

19

0.63 0.01

1298 18.82

78.08

Standard

0.24

0.94