Prediction of the Strength Characteristics of Limestones in the Sanandaj – Sirjan Zone Using Statistical Methods and Neural Network

Rahimi Shahid, Mojtaba; Lashkaripour, Gholam Reza; Hafezi Moghaddas, Naser

doi:10.22034/JEG.2025.19.2.1018062

Volume 19, Issue 2 (Summer 2025) 2025, 19(2): 290-327 | Back to browse issues page

‎ 10.22034/JEG.2025.19.2.1018062

Research code: 3/56943

Mendeley

Zotero

RefWorks

Rahimi Shahid M, Lashkaripour G R, Hafezi Moghaddas N. Prediction of the Strength Characteristics of Limestones in the Sanandaj – Sirjan Zone Using Statistical Methods and Neural Network. Journal of Engineering Geology 2025; 19 (2) :290-327
URL: http://jeg.khu.ac.ir/article-1-3172-en.html

Prediction of the Strength Characteristics of Limestones in the Sanandaj – Sirjan Zone Using Statistical Methods and Neural Network

Mojtaba Rahimi Shahid¹

, Gholam Reza Lashkaripour²

, Naser Hafezi Moghaddas¹

1- Ferdowsi University of Mashhad, Mashhad, Iran
2- Ferdowsi University of Mashhad, Mashhad, Iran , Lashkaripour@um.ac.ir

Abstract: (599 Views)

The Sanandaj–Sirjan Structural-Sedimentary Zone is one of the most important geological regions in Iran. The limestone formations in this area play a key role in civil engineering and mining projects. Knowing the precise mechanical properties of these rocks, especially the uniaxial compressive strength (UCS dry) and dry point load index (Is₅₀-dry), is essential for safely and economically designing structures. Because direct testing methods are costly and time-consuming, this study uses indirect modeling techniques, such as regression and neural networks, to predict these properties. First, a comprehensive database was compiled by collecting the physical, mechanical, dynamic, and chemical data of limestone samples from the region. Then, univariate, bivariate, and multivariate regression analyses were conducted to extract statistical relationships among the variables. Finally, multilayer perceptron neural network models with various architectures based on the Levenberg–Marquardt learning algorithm were developed. The comparison results of the model performance indicated that neural networks, due to their ability to identify complex and nonlinear relationships between parameters, provide more accurate predictions of the limestone mechanical properties compared to statistical models. A comparison of the correlation coefficients of multivariate regression equations and neural network models showed that, overall, using neural network models improves the accuracy of UCS Dry predictions by 14.89% and the Is ₅₀-Dry predictions by 4.70%. The results show that predicting UCS Dry in the presence of Is ₅₀-Dry among the input parameters has a significant impact on improving the accuracy of the models. For example, the model with the inputs Is ₅₀-Dry, SH, γ Dry and n showed very good performance. For predicting Is ₅₀-Dry, the models that included the parameters SDI1 and BI Dry as inputs also performed very well. The application of these models can contribute to cost reduction, increased speed of rock engineering studies, and improved safety in civil projects.

Keywords: Multi Layer Perceptron, Engineering Geological Characteristics, Regression, Point load index, Uniaxial compressive strength.

Full-Text [PDF 3951 kb] (37 Downloads)

Type of Study: Original Research | Subject: Engineering Geology
Received: 2025/07/6 | Accepted: 2025/09/3 | Published: 2025/09/22