PSI - Issue 70

Santhosh Kumar N V et al. / Procedia Structural Integrity 70 (2025) 440–446

443

Table 1. Statistical parameters appertaining to each of the input attributes Attribute Unit

Min.

Max. 412.11 235.27 229.27 1429.69 413.98

Mean 305.42 54.14 75.37 1304.50 330.34

Std. Dev.

Lime Ggbs

Kg/m 3 Kg/m 3 Kg/m 3 Kg/m 3 Kg/m 3 Days

210.85

45.98 80.42 72.52 53.12 30.26 23.99 1.78

0.00 0.00

Inputs

Redmud

Fine Aggregate

1130.66 264.11

Water

Age

1.00 0.32

90.00 9.90

24.39 2.29

Output

Compressive Strength

MPa

The main goal of this research is to develop accurate and reliable ML models that predict compressive strength, they will ultimately provide knowledge that aids design of compressive strength without the need for additional laboratory testing and assist in further advanced materials optimization. The study employs supervised learning algorithms, which include Random Forest (RF), Support Vector Machine (SVM), XGBoost, K-Nearest Neighbours (KNN), and Multiple Linear ones (MLR), to find the best model performance. The authors partitioned the data into a training and testing subset, structures like feature importance and cross-validation were performed to improve both prediction and modelling accuracy. The models’ ML findings will be evaluated against Mean Squared Error (MSE), Mean Absolute Error (MAE), and Adjusted R-squared statistics (Chen & Guestrin, 2016, Chouhan et al. 2020). 3.2 Model evaluation metrics Proper evaluation of machine learning models is critical to determine how well these models can predict outcomes. In the present study, we applied three performance measures - Mean Squared Error (MSE), Mean Absolute Error (MAE), and Adjusted R-Squared (R²) - to evaluate the accuracy of each of the developed models for estimating the compressive strength of lime mortar. Smaller MSE and MAE values reflect greater model accuracy, and higher Adjusted R-Squared (R²) indicates a stronger predictive relationship. 3.2.1 Mean squared error (Mse) MSE is the average squared difference between actual and predicted values; a lower MSE is a more accurate model. ( ) 2 1 1 n i i i MSE y y n = = −  (1)

3.2.2 Mean Absolute Error (Mae)

MAE has a similar intent to MSE, but MAE measures absolute differences between actual and predicted values. As a result, this is an easier and more intuitive value to interpret in terms of prediction error.

1 n n = i

i − 

(2)

MAE

y y

=

i

1

3.2.3 Adjusted R-Squared (R²): Adjusted R-Squared (R²) gives an estimate of R-Squared (R²) that is adjusted for the number of predictors in a model; this seeks to avoid over estimating performance, by imposing a penalty for using more models than necessary. R² values can be viewed relative to 0 or 1 model fit. The closer value to 1 indicates a stronger model fit. ( ) ( ) 2 2 1 1 _ 1 1 R n R adj n k   − −   = −   − −   (3)