PSI - Issue 70

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

398

3. An Evaluation of the Results 3. Results of the Decision Tree Model Analysis

DT model results for GPC, CS are shown in Figure 5. compares experimental and predicted findings. The DT output was accurate, and experimental findings matched forecasts. The DT model accurately predicts the chosen concrete's compressive strength with an R2 of 0.88. Figure 9b shows the disparities between DT model predictions, expected results, and actual findings. Minimum, mean, and highest values were 0.50, 4.14, and 29.68 MPa. Distribution of results: 9.6% had a pressure below 43.8% between 21.9%, 16.4% between 5 and 10 MPa, and 8.2% over 10 MPa. The data' dispersion suggests a robust DT model.

Fig. 4. Decision tree model- experimental-predicted connection

Fig. 5. Random forest model result: Relationship

between

experimental and model results

3.2. Random Forest Model Shows in fig. 5 the RF model's 0.95 R2. Figure, predicted, and RF model dispersion. Minimum error was 0.40 MPa, average was 2.59 MPa, and maximum was 13.42 MPa. The dispersion of error values was 23.29%, 50.68% between 2 and 4 MPa, and 26.03% above 4 MPa. The distribution shows the RF model's improved prediction. Single SML algorithms were less accurate than ensembles. 3.3. Validation The confirmation of these models was achieved through statistical testing and KFCV. The majority of models undergo validation through the KFCV method (Pilehvar et al., 2018) which randomly segments similar data into 10 groups. Nine trains and one validates the selected model, as illustrated. The models underwent training using 80% of the database, while the remaining 20% was utilised for testing. A high R2 along with low MAE and RMSE indicates that the model is more effective for making predictions. The procedure needs to be performed ten times to achieve success. This comprehensive approach enhances model forecasting. Table 2 shows that MSE and RMSE errors were utilised to evaluate all ML models. The results of these tests indicated that the BR model demonstrated enhanced accuracy due to its reduced reading error. Statistics evaluated the prediction performance of the models. Table 2. the statistical results obtained from the models that were used.

Table 2. Results obtained from KFCV for each and every model that was used. SML Technique MAE

RMSE

Decision tree(DT) Random forest(RF)

4.136 2.585

6.256 3.702