PSI - Issue 70

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

383

2.3. Support Vector Machine (SVM) Supervised learning methods, such as SVM, are employed for classification and regression evaluation. SVM techniques represent samples as point in space, arranged to separate the patterns of various classifications using a vector (line/plane) that maintains a significant margin. illustrates how additional examples are integrated into that similar area and classified by vector side SVM model. This model evaluates material strength by taking into account various factors. The parameters of the SVM model were established through optimization. 3. Analysis of Results 3.1. Support Vector Machine Model (SVM) Fig. 2 presents the results of GPC's CS SVM analysis and a comparison between experimental and projected data. The SVM approach yielded precise outcomes, showing only a slight discrepancy between the actual and predicted results. SVM provides a more precise prediction of GPC CS, achieving an R2 value of 0.93. Fig. 3 illustrates the scattering of experimental, predicted, and divergence errors for the SVM model. Upon analyzing the error values, the minimum, average, and maximum recorded were 0.20, respectively. Furthermore, the fraction diffusion of divergence values indicated that 37% fell below 3 MPa, 37% ranged from 3 to 5 exceeded 5 MPa. Furthermore, the variation of divergent values indicates that SVM effectively predicted GPC CS.

Fig. 2. SVM Model- connection between real and expected

Fig. 3. distribution of experimental and predicted results.

4. Conclusions This study evaluated the compressive strength of geopolymer composites using both individual and ensemble machine learning techniques. SVM is used as a single approach. This investigation produced the following outcomes: The model surpassed the performance of individual machine learning methods in predicting the CS of GPCs. The R2 values for the SVM models were 0.98 respectively. All techniques employed yielded precise results with slight variation from experimental findings. The method excelled in predicting the CS of GPC. The accuracy of the ML model was demonstrated through a reduction in errors and an improved R2. The model demonstrated superior performance compared to other models, with GGBS identified as the most significant input feature, showing a positive correlation with GPC's CS, as revealed by SHAP analysis. This study will aid the construction sector in developing efficient and economical material forecasting tools. Furthermore, advocating for sustainable construction will hasten the integration of GPC within the building sector. Additional experimental parameters such as sample geometry, strain rate, and temperature effects can be explored. Random forest, bagging, and boosting can also be employed to assess the accuracy of results. Acknowledgements