PSI - Issue 70

Abdul Musavvir et al. / Procedia Structural Integrity 70 (2025) 432–439

437

4. Validation Validation of research is necessary prevent misleading conclusions, and support reliable decision-making in scientific and practical applications. Mix design of M70 grade concrete is made for HSC. The mix design was made into eight combinations with replacement of the cement with supplementary cementitious material from industrial wastes which poses cementitious properties. The estimated cement content is 416 kg/m 3 of OPC53 grade and 2% super plasticizer (SP 430) is used. The W/c ratio used is 0.25 and volume of fine aggregate and coarse aggregate in ratio to cement are 1.75 and 3.55. The supplementary cementitious material from industrial wastes like Fly Ash, Ground Granulated Blast Furnace Slag, Silica Fumes and Nano silica were used in replacement of the cement.

Table 3. Experimentation Mix Combinations Mix No

Proportion

MIX 1 MIX 2 MIX 3 MIX 4 MIX 5 MIX 6 MIX 7 MIX 8

90% Cement + 10% Fly ash 90% Cement + 10% GGBS

90% Cement + 10% Silica Fumes

90% Cement + Nano Silica

80% Cement + 10% Fly ash + 10% GGBS 80% Cement + 10% GGBS + 10% Silica Fumes 80% Cement + 10% Silica Fumes + 10% Nano Silica 80% Cement + 10% Nano Silica + 10% Fly ash

Ground Granulated Blast Furnace Slag (GGBS)

5. Result and Discussion The ML models have been developed using different algorithms like Linear Regression, Decision Tree Regression, Random Forest and XGBoost Regression. Out of the models developed the XGBoost Regression shows good performance in evaluation using R2 Score, Mean Squared Error (MSE), Mean Absolute Error (MAE), Root Mean Squared Error (RMSE) and Relative Mean Error (RME). The discussion is more focus on XGBoost Regression model in general. • XGBoost excels due to its efficient implementation of gradient boosting, which combines multiple weak trees to create a strong predictive model. Built-in regularisation to prevent overfitting, effective handling of missing values, and faster training using parallel processing. Additionally, XGBoost’s ability to optimi se performance through hyperparameter tuning and its adaptability to various data types make it among the best models in ML. • The Model shows R2 score of 0.85 which denotes the huge variance in data. Initially the R2 score was 0.83 which had a dataset of 200 data. While, after the collection of dataset of 155 the R2 score increased. • The Model has higher accuracy when focused on 50-80 MPa due to majority of data is from the range of 50 80 MPa, its accuracy slightly decreases when the prediction goes beyond 80 MPa due to lack of data in the range of 80-100 MPa. To increase the accuracy in that range data must be added in that range specifically. • Similarly the accuracy of model is high (>96%) while making prediction for Ground Granulated Blast Furnace Slag and Fly Ash due to large data availability of these parameters other than concrete and super plasticizer. But, relatively less for predicting Silica fumes and Nano Silica as it has small amount of data. • The Linear Regression and Decision Tree Regression were not performing well due to the complexity of data. These models works good for linear data. As the dataset consist of many non-linearity due to supplementary cementitious industrial wastes. (Mohammad Mohtasham, 2023) • The accuracy of XGBoost model can be increased when further data is added and trained with consideration to the above discussed areas which affect the accuracy of prediction.