INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue IX September 2025
Page 2738
Logging Data-Driven Geomechanical Parameter Estimation Using
Advanced Machine Learning Techniques
1
Osaki Lawson-Jack.,
2
Oghonyon Rorome
1
Department of Physics and Geology, Federal University Otuoke, Bayelsa State, Nigeria
2
Department of Geology, University of Port Harcourt, Rivers State, Nigeria
DOI: https://doi.org/10.51244/IJRSI.2025.120800242
Received: 31 July 2025; Accepted: 13 Aug 2025; Published: 02 October 2025
ABSTRACT
The most commonly used methods of conventional geomechanical parameters estimation which rely on costly,
sparse laboratory tests and empirical correlations based on just a few well logs are linked to uncertainties and
spatial gaps. This study reveals an innovative data-driven model, which incorporates Advanced Machine
Learning techniques to precisely and efficiently estimate key geomechanical properties based directly on
collected well-logging data. The techniques include, Deep Learning Architecture (DL), Deep Neural Network
(DNN) and Artificial Neural Network (ANN). The machine learning application ensures a huge boost to
yielding high prediction accuracies and that of running continuous and high-resolution profiles of
geomechanical properties along the wellbore. This method is fast, and has low-cost geomechanical
characterization that is vital to optimal drilling, hydraulic fracturing design, reservoir management, and
subsurface integrity assessment, resulting in improved operating safety and efficiency. The estimated
geomechanical parameters include elastic properties (young’s modulus and poisson’s ratio) and rock’s strength
(unconfined compressive stress), while the artificial neutral network technique was applied to estimate the
geomechanical parameters in the oil wells of Akata, Agbada and Benin Formations in Bonny Island, Rivers
State.
Keywords: Logging Data-Driven, Geomechanical Parameter Estimation, Advanced Machine Learning,
Comparative Analysis of Techniques, Geomechanical Properties.
INTRODUCTION
Mechanical and petro-physical properties of rocks are characterized by their textural properties. To a high
extent, such parameters determine the stability of the rock mass. The capacity of assessing both short- and
long-term rock behaviors according to the interaction between distinct parameters of rock texture,
petrophysical and mechanical properties are thus highly instrumental to a number of geoengineering materials
(Askaripour et al., 2022). Lin et al., (2021) opined that, how the properties of rock affect the mechanism of
electromagnetic radiation (EMR) phenomenon of the process of rock fracture is an issue that is significant to
study in solid mechanics and earthquake forecasting. According to Yan et al., (2020) Rock anisotropy is an
intrinsic property of natural rock mass, and layered rock has the most significant effect on the stress
distribution and deformation of a rock mass. With the development of rock mechanics theory and constitutive
theory, the study of rock anisotropy has become one of the focuses and hotspots in the field of rock mechanics.
LITERATURE REVIEW
Few studies have been carried out on fluid-rock interaction like adsorption, precipitation, fines migration, and
wetting properties in porous media utilizing EOR (Enhanced oil recovery) fluids, however, the precise
mechanism of these fluids that occurs during the EOR application of rock remains uncertainly revealed.
Depending on a set of parameters, these fluid- rock interactions determine the scope and consequences of these
interactions on EOR. These factors are type of fluid injected and the composition of chemicals, the type of
rock and mineralogical composition, brine PH, brine salinity and composition. Furthermore, all the methods of
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue IX September 2025
Page 2739
quantification of fluid-rock interactions possess certain drawbacks in terms of their application, measurement
range, or the level of uncertainty (Isah et al., 2022).
Thin-section identification in rocks is a very crucial geological exploration instrument in interpreting and
identifying the structure of the earth. It also turns out to be a significant assessment technique of oil and gas
exploration and development. It can focus on the identification of petrological properties of the reservoirs, the
type of diagenesis, distinction of the reservoir cave space and pore features. Those properties of physical
nature and sedimentary environment of the reservoir have to be comprehended, the parameters desired of
reservoir attained, oil and gas development plan and reservoir calculation has to be made. The conventional
thin-section identification technology had an over one-hundred-year history and relied mainly on the visual
identification of the geological experts with the help of the optical microscope, and was plagued with the
shortcomings of strong subjectivity, high experience dependence and heavy work-load, long-identification-
cycle, and inability to reach complete and accurate quantification (Liu et al., 2022).
Geomechanical data are never adequate in quantity, proportion, precision, and accuracy to be utilized in
design. This stems out of the fact that rock masses are naturally complex and variable in all scales.
Geomechanical properties of the rock masses are not completely random in theory. Since rocks were created
and constantly altered with multiple complicated processes, which lead to physical heterogeneity resulting to
differences in the values of measured physical properties, even in a single rock type. Moreover, the natural
fractures exist and this leads to the existence of spatial and regional differences in rock mass property, i.e.
natural fractures bring about spatial and regional variation (Małkowski et al., 2021).
In order to develop a successful geomechanical characterization of the rock masses, Heidarzadeh et al., (2021)
reported a suitable interpretation on the rock masses lithological heterogeneity ought to be achieved where
both the geological and geomechanical data would be considered. To better explain the reliability and
usefulness of geological surveys in application to the field of rock mechanics, a geomechanical
characterization study is made on the heterogeneous Niobec Mine (Quebec, Canada) by taking into account the
nature of the various lithological units identified in the mass. The resulting outcomes of the past field and
laboratory testing campaigns, in terms of lithological units, became part of determining the variability related
to the intact rock geomechanical parameters of the various current lithological units.
METHODOLOGY
A. Geomechanical Parameters Estimation
1. Elastic Properties: Elastic properties of rocks can be determined through laboratory measurement and well
log data parameters (Davy et al., 2018). The elastic properties for this research was determined through well
log data, and the parameters include dynamic young’s modulus and dynamic poisson’s ratio.
1. Young Modulus
It measures the rock’s stiffness that makes it resistant to deformation under stress, especially during drilling
operations (Mahdi & Alrazzaq, (2023).
…………………………………………..1
 



………………………..2
The Young’s modulus of each types of rock are presented in table 1 below
Table I: Young’s Modulus of Rock Types: Source from Małkowski et al., (2021).
Rock Type
Young’s Modulus (GPa)
Level
Effect
Soft Shale
0.5-5
Low
Easy deformation
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue IX September 2025
Page 2740
Sandstone
10-70
Medium
Oil or gas extraction fracability
Limestone
30-80
High
Tendency of cracking
Granite
50-100
Very High
Strong and rigid
Salt
10-30
Medium
Creeping material
Poisson’s Ratio (ν): It measures lateral strain against the axial strain of the rock (Lutz & Zimmerman, 2021).
The poisson’s ratio data for different rock types are presented in table 2 below;



……………………………………………………………………3
Table II. Poisson’s Ratio Data of Rock Types: Source from Małkowski et al., (2021).
Rock Type
Level
Effect
Soft Shale
High
High lateral expansion
Sandstone
Medium
Forecasting of stress anistrophy
Coal
Low
Low lateral strain
Salt
Very High
High ductility property
2. Rock Strength: The rock strength determines the force applied during drilling operations at the oil
wells. Therefore, it is expedient to estimate the stress value of the rock (Kalantari et al., 2022).
1. Unconfined Compressive Stress
It measures maximum axial stress before failure (Li & Yang, 2024). The unconfined comprehensive stress data
is presented in table 3 below;
Table III: Unconfined Compressive Stress Data of Rock Types (Source from Lin et al., (2021).
Rock Type
UCS (MPa)
Level
Implication
Weak Shale
0.25-0.40
Low
Easy to collapse
Sandstone
0.10-0.30
Medium
Too hard to drill
Limestone
0.10-0.15
High
High rigidity
Granite
0.35-0.45
Very High
Little drillable condition
Chalk
1-15
Extremely weak
Instability
2. Friction Angle (φ)
It measures internal shear resistance e.g angle of failure in Mohr-Coulomb theory (Zoorabadi & Muralha,
2025). The friction angle data for rocks at oil wells are presented in table 4 below;
θ = tan⁻¹(μ)…………………………………………………………………………4
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue IX September 2025
Page 2741
Table IV. Friction Angle Data of Rock Types (Source from Zoorabadi & Muralha, 2025).
Rock Type
Friction Angle (φ) (
o
)
Level
Geomechanics Responsibility
Clay
10-20
Low
High landslide
Sandstone
25-40
Medium
Monitors the pressure of shear failure
Conglomerate
35-45
High
High stability
Fractured Rock
Less than 15
Very Low
Risk of deficiency activation
B. Estimate Comparison of Young’s Modulus, Poisson’s Ratio and Unconfined Compressive Stress
The comparison of the geomechanical parameters data determined through well log data are presented in figure
1 below;
Fig. 1. Geomechanical Parameters Data Comparison determined from well log data (Source from Sanei et al.,
2023).
C. Advanced Machine Learning Techniques
1. Deep Learning Architecture: The major difference between the conventional machine learning model and
advanced learning model is its automatic learning process which makes deep learning suitable for wide range
of applications (Endo, 2023). Sewak et al., (2020) opined that, based on the applications and types of neural
networks, deep learning architecture is classified into three major classes, as presented in the figure 2 below;
Fig. 2. Classes of Deep Learning Architecture in Machine Learning (Source from Smys et al., 2020).
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue IX September 2025
Page 2742
Generative Architecture: They are generally called unsupervised feature learning model, because they are
generative in nature. The data labels are not taken into consideration in this strategy. Such kind of architecture
is developed when there is a little space, such models learn the lower level of the data and offers the required
solutions to the hard network, because data is trained to work without relying on other layers (Caetano et al.,
2020).
Discriminative Architectures: In the processing of information and signals, discriminative architectures are
mostly dominating. The deep structures with conditional random fields have been developed whose output at
one level (random field) of the lower part becomes stacked upon the original input data that is on upper layer
(Smys et al., 2020). According to Bhatt et al., (2021) language processing uses discriminative architectures and
identification apps. Through these apps, HMM (Hidden Markov Model Tools) are witnessed through the
activities of the hidden layers in form of different combinations that make up a discriminative architecture.
Hybrid Architecture: Hybrid architecture has both discriminative and generative process. The generative
parts are utilized and combined with discriminative parts in order to achieve the last solution. The generative
models are applied to solve non-linear parametric problems which decreases the initialization issues. Also
generative models have regularized control features making the system simple (Yang et al., 2022). Liu and
Abbeel, (2020) illustrated an example of how Deep Neural Network (DNN) is a recognized hybrid framework
in which the generative framework of deep network is employed. Deep Neural Network is developed by
modifying the belief network based on the discriminative architecture in training process.
2. Deep Neural Network: Deep Neural Networks (DNNs) have transformed the study of rocks in geology by
automating the process of pattern recognition of intricate geometries, offer higher accuracy, and guarantee
timely analyses of geological engineering, resource prospecting and hazard mitigation (Samek et al., 2021). Li
et al., (2023) opined that, mathematically, more complex deep learning strategies like deep neural networks
(DNN) have been formulated to explore multi-variable systems which have shown similar, and even better
performance than human experts. The deep and shallow neutral network layers are presented in figure 3 below;
Fig. 3. Deep and Shallow Layers of Deep Neural Networks (Source from Azarafza et al., 2022).
3. Artificial Neural Networks: Artificial Neural Networks have the ability to convert raw well logs to high-
resolution profiles of rock properties that describe geomechanics by acquiring an elaborate non-linear
connection (Qiang et al., 2020). Millán et al., (2021) noted that, this involves log measurements as well as rock
property measurements that have been checked in the laboratory or using cores. The networks applicable to
ANNs are presented in table 5 below;
Table V. Artificial Neural Networks : Source from Millán et al., (2021): Qiang et al., (2020).
Network Type
Structure
Use Case
Multilayer Perceptron (MLP)
3–8 fully connected (dense) layers
Predicting UCS or E-static from 5–7
logs.
1D Convolutional Neutral
Network (CNN)
Convolutional layers + pooling for depth
patterns
Detecting thin-bed effects on stress
(σ).
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue IX September 2025
Page 2743
Hybrid CNN-MLP
CNN extracts spatial features → MLP
maps to outputs
Pore pressure prediction from log
sequences.
Physics-Informed NN (PINN)
Custom loss enforcing rock physics rules.
Stress estimation obeying Hooke’s
law.
ANN can be best described through the following;
Hyperparameters
According to, Kadhim et al., (2022) Hyperparameters regulate the learning of the ANN. The major examples
of hyperparameters include;
Hidden layer/neurons: Determines the complexity of the model. 1–3 hidden layers (with 10–50 neurons
each) typically balance accuracy and efficiency. Although geomechanics implementations often use three
hidden layers containing 10 to 50 neurons.
Learning rate: Controls the step length in the optimization (e.g., Adam). Noise in log descriptions of data
causes no overshooting of minima at 0.001 or 0.01.
Regularization (L2/dropout): The large weights are punished (L2) or the neurons are randomly removed
(dropout) to mitigate over-fitting.
ANN Architecture
ANN architecture encompass input, hidden and output layers, as presented in figure 4 below. Madhiarasan and
Louzazni, (2022), described the layers as:
Input layer: Takes the normalized logs (e.g GR, RHOB, DTC).
Hidden layers: Use summation of network weights and a non-linear activation function (e.g. ReLU), to learn
features.
Output layer: It provides approximations (e.g. UCS, Poisson Ratio). It can include image logs which are a
type of hybrid architecture (e.g., CNN-MLP).
Figure 4: Artificial Neural Networks Architecture Design (Source from Azarafza et al., 2022).
Training and Validation Approach
Training: 7080 per cent depth-indexed core-log pairs are training to minimize weights through
backpropagation. Convergence stabilizes when the training is done in mini-batch (Livieris, 2018).
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue IX September 2025
Page 2744
Validation: Hyperparameters are tuned and early stopping applied in case of stagnation on the loss with 15-20
percent of the data. Robustness is guaranteed by K-fold cross-validation between the wells (Li t al., 2022).
Performance Metrics
Erickson & Kitamura, (2021) described the major performance metrics of ANN as,
RMSE (Root Mean Squared Error): Mainly used to measure regression e.g. RMSE < 1.5 GPa to Youngs
Modulus.
R
2
(Coefficient of determination): Represents the proportion of variance, the higher the better (>0.90).
MAE (Mean Absolute Error): It is very receptive to outliers in fundamental measurements.
Overfitting and Underfitting
Salman and Liu, (2019) revealed that,
Overfitting: model learns the noise in the training data and overfit with new wells. Reduce it through dropout,
L2 regularization, and network size-reduction.
Underfitting: oversimplified structure can overlook important log-parameter correlations. Train more layer
neurons or more iterations.
D. Comparative Analysis of Techniques
The Artificial Neural Networks and Deep Learning Architectures are comparable in that they both imitate the
functioning of biological neurons, but differ fundamentally in their size, complexity, and capacities
(Montesinos López et al., 2022). Saikia et al., (2020) noted that, there are shallow ANNs that are at most 2
layers deep and are known to excel at easier problems, such as a regression or analysis of simple logs. Their
lightweight design fits the small data cases but it is not designed to handle raw, and high-dimensional logging
data. In Deep Learning (DL) Architectures, numerous hidden layers, e.g. CNNs (Convolutional Neural
Networks), and RNNs (Recurrent Neural Networks) are stacked, allowing to automatically learn features (Guo
et al., 2023). Kufel et al., (2023) noted that, Deep Learning lives in large amounts of data (terabytes logs) and
yet requires GPUs, however, it excels ANNs on complex applications such as 3D prediction of reservoir
properties. Compared to ANNs which are manual transmission (controlled but constrained), DL is more self-
driving (autonomous but resource-intensive).
E. Research Study Area
The Bonny Island is located in the Rivers State and is at the center of Niger Delta petroleum system, which is a
world-class hydrocarbon province that is typified by its complex, and prograding deltaic sequences (Obasohan
et al., 2021). Bankole et al., (2014) noted that, the main geology in the subsurface is dominated by Agbada
Formation which was a critical period that contained interbedded sandstones and shales that were laid down in
a delta-front to shallow marine landscape in the Miocene to Pliocene. These sand bodies are the main
hydrocarbon reservoirs but they are highly heterogeneous as to their thickness, grain size, as well as the clay
content because the depositional environments were changing frequently.
The Benin Formation is a generally unconsolidated continental sand in which overpressured Akata Formation
shales becomes the main regional source-rock and seal. Such a geological environment poses difficult
prospects to geomechanical modeling. The stratigraphy is dissected by growth faults which form
compartmentalized reservoirs and with abrupt change in stress orientation and magnitude. They are
interbedded sequences of sand-shale that are mechanically anisotropic with sands easily compacted and shales
easily swelling or failing. The unusually fast rates of sedimentation have resulted in high and abnormal pore
pressures especially in and around fault planes and in deeper shales units, and extremely limited drilling mud
weight windows (Diab et al., 2023).
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue IX September 2025
Page 2745
Due to the fact that the wellbore instability risk in addition to sanding and fault reactivation threats are acute in
Bonny, high-fidelity serves the geomechanical model required environment. This is what made its complexity
ideal of machine learning (ML) methods. ML algorithms have a potential to uncover concealed regularities in
this data, combining measurements that lack spatial consistency into an approximation of spatially variable
geomechanical parameters more reliable than empirical correlations, and eventually optimizing drilling safety
and reservoir management in this high stakes deltaic environment (Ogoro, 2014). The Bonny Island map is
presented in figure 5 below;
Fig. 5. Map Showing Bonny Island (Source from Obasohan et al., 2021).
RESULTS AND DISCUSSION
F. Geomechanical Parameters Estimation
The estimated Young’s Modulus data of the Akata, Agbada and Benin formations in the Bonny Island are
presented in table 6 below;
Table VI: Young’s Modulus Data of the Akata, Agbada and Benin Formations
Depths
Akata Formation
Agbada Formation
Benin Formation
Shallow Depths
2-15 MPa
100-500 Mpa
10-50 Mpa
Intermediate Depths
10-50 Mpa
500-2,000 MPa
50-150 Mpa
Greater Depths
30-100 Mpa
› 2,000 Mpa
150-400 Mpa
The estimated Poisson’s ratio data of the Akata, Agbada and Benin formations in the Bonny Island are
presented in table 7 below;
Table VII: Poisson’s Ratio Data of the Akata, Agbada and Benin Formations
Conditions
Akata Formation
Agbada Formation
Benin Formation
Undrained
0.45-0.49
0.30-0.40
Short term loading – 0.45-0.49
0.35-0.45
Short term loading – 0.45-0.49
Drained
0.30-0.40
0.25-0.35
Long term loading – 0.30-0.40
0.25-0.35
Long term loading – 0.30-0.40
Dynamic
0.40-0.48
0.33-0.38
Long data – 0.40-0.48
0.33-0.42
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue IX September 2025
Page 2746
The estimated Unconfined Compressive Stress data of the Akata, Agbada and Benin formations in the Bonny
Island are presented in table 8 below;
Table VIII. Unconfined Compressive Stress Data of the Akata, Agbada and Benin Formations
Depths
Akata Formation
Agbada Formation
Benin Formation
Shallow Depths
5-20 kPa
2-10 MPa
10-30 kPa
Intermediate Depths
10-50 kPa
10-40 MPa
30-100 kPa
Greater Depths
50-200 kPa
40-150+ MPa
100-300 kPa
G. Advanced Machine Learning Models
The well logged geomechanical parameters were estimated through the artificial neutral network, specifically
the Multilayer Perceptron (MLP) technique, as presented in the figures below;
Fig. 6. Young Modulus Parameter of Multilayer Perceptron (MLP) Technique
Fig. 7. Poisson’s Ratio Parameter of Multilayer Perceptron (MLP) Technique
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue IX September 2025
Page 2747
Fig. 8. UCS Parameter of Multilayer Perceptron (MLP) Technique
CONCLUSION
High-precision geomechanical parameters estimation of normalized logs using advanced machine learning
techniques such as, DL, DNN and ANN was realized, which proved to be more effective than empirical
methods. With these systems in real-time, dynamic wellbore stability alerts could be achieved with less than 50
milliseconds delay. However, the discussed models are data dependent and need quality logs or core samples
to train on. Violation of physics is possible without any limit, and due to the computational cost, it is difficult
to deploy edges in complicated 3D environments. This study has contributed to end-to-end workflows so as to
transform routine logs to lab-grade mechanical properties at reduced costs through fewer tests. The hybrid
architectures helped to fill the gap between data scarcity and physical realism, which provided field-deployable
solutions to proactive geomechanical management efforts.
ACKNOWLEDGEMENT
The authors of this paper express gratitude to the management of Monipulo Petroleum Limited for supplying
the essential data and software utilised in conducting this research.
REFERENCES
1. Askaripour, M., Saeidi, A., Mercier-Langevin, P., & Rouleau, A. (2022). A review of relationship
between texture characteristic and mechanical properties of rock. Geotechnics, 2(1), 262-296.
2. Azarafza, M., Hajialilue Bonab, M., & Derakhshani, R. (2022). A deep learning method for the
prediction of the index mechanical properties and strength parameters of marlstone. Materials, 15(19),
6899.
3. Sanei, M., Ramezanzadeh, A., & Delavar, M. R. (2023). Applied machine learning-based models for
predicting the geomechanical parameters using logging data. Journal of Petroleum Exploration and
Production Technology, 13(12), 2363-2385.
4. Bankole, S. I., Schrank, E., & Osterloff, P. L. (2014). Palynostratigraphy, palaeoclimates and
palaeodepositional environments of the Miocene aged Agbada Formation in the Niger Delta,
Nigeria. Journal of African Earth Sciences, 95, 41-62.
5. Bhatt, D., Patel, C., Talsania, H., Patel, J., Vaghela, R., Pandya, S., & Ghayvat, H. (2021). CNN
variants for computer vision: History, architecture, application, challenges and future
scope. Electronics, 10(20), 2470.
6. Caetano, I., Santos, L., & Leitão, A. (2020). Computational design in architecture: Defining parametric,
generative, and algorithmic design. Frontiers of Architectural Research, 9(2), 287-300.
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue IX September 2025
Page 2748
7. Davy, P., Darcel, C., Le Goc, R., & Mas Ivars, D. (2018). Elastic properties of fractured rock masses
with frictional properties and power law fracture size distributions. Journal of Geophysical Research:
Solid Earth, 123(8), 6521-6539. https://doi.org/10.1029/2017JB015329
8. Diab, A. I., Sanuade, O., & Radwan, A. E. (2023). An integrated source rock potential, sequence
stratigraphy, and petroleum geology of (Agbada-Akata) sediment succession, Niger delta: application
of well logs aided by 3D seismic and basin modeling. Journal of Petroleum Exploration and Production
Technology, 13(1), 237-257.
9. Endo, T. (2023). Analysis of Conventional Feature Learning Algorithms and Advanced Deep Learning
Models. Journal of Robotics Spectrum, 1, 001-012.
10. Erickson, B. J., & Kitamura, F. (2021). Magician’s corner: 9. Performance metrics for machine
learning models. Radiology: Artificial Intelligence, 3(3), e200126.
https://doi.org/10.1148/ryai.2021200126
11. Guo, Q., He, Z., & Wang, Z. (2023). Prediction of monthly average and extreme atmospheric
temperatures in Zhengzhou based on artificial neural network and deep learning models. Frontiers in
Forests and Global Change, 6, 1249300.
12. Heidarzadeh, S., Saeidi, A., Lavoie, C., & Rouleau, A. (2021). Geomechanical characterization of a
heterogenous rock mass using geological and laboratory test results: a case study of the Niobec Mine,
Quebec (Canada). SN Applied Sciences, 3, 1-20.
13. Isah, A., Arif, M., Hassan, A., Mahmoud, M., & Iglauer, S. (2022). Fluid–rock interactions and its
implications on EOR: Critical analysis, experimental techniques and knowledge gaps. Energy
Reports, 8, 6355-6395.
14. Kadhim, Z. S., Abdullah, H. S., & Ghathwan, K. I. (2022). Artificial Neural Network Hyperparameters
Optimization: A Survey. Int. J. Online Biomed. Eng., 18(15), 59-87.
15. Kalantari, S., Baghbanan, A., & Hashemalhosseini, H. (2019). An analytical model for estimating rock
strength parameters from small-scale drilling data. Journal of Rock Mechanics and Geotechnical
Engineering, 11(1), 135-145.
16. Kufel, J., Bargieł-Łączek, K., Kocot, S., Koźlik, M., Bartnikowska, W., Janik, M., & Gruszczyńska, K.
(2023). What is machine learning, artificial neural networks and deep learning?—Examples of practical
applications in medicine. Diagnostics, 13(15), 2582.
17. Li, H., & Yang, M. (2024). Study on unconfined compressive strength and deformation characteristics
of chlorine saline soil. Scientific Reports, 14(1), 1478.
18. Li, S., Hu, J., Amann, F., Li, L., Liu, H., Shi, S., & Hamdi, P. (2022). A multifunctional rock testing
system for rock failure analysis under different stress states: Development and application. Journal of
Rock Mechanics and Geotechnical Engineering, 14(5), 1531-1544.
https://doi.org/10.1016/j.jrmge.2021.12.017
19. Li, Z., Li, H., & Meng, L. (2023). Model compression for deep neural networks: A
survey. Computers, 12(3), 60. https://doi.org/10.3390/computers12030060
20. Lin, P., Wei, P., Wang, C., Kang, S., & Wang, X. (2021). Effect of rock mechanical properties on
electromagnetic radiation mechanism of rock fracturing. Journal of Rock Mechanics and Geotechnical
Engineering, 13(4), 798-810.
21. Liu, H., & Abbeel, P. (2020). Hybrid discriminative-generative training via contrastive learning. arXiv
preprint arXiv:2007.09070. https://doi.org/10.48550/arXiv.2007.09070
22. Liu, H., Ren, Y. L., Li, X., Hu, Y. X., Wu, J. P., Li, B., & Fang, W. K. (2022). Rock thin-section
analysis and identification based on artificial intelligent technique. Petroleum Science, 19(4), 1605-
1621.
23. Livieris, I. E. (2018). Improving the classification efficiency of an ANN utilizing a new training
methodology. In Informatics (Vol. 6, No. 1, p. 1). MDPI. https://doi.org/10.3390/informatics6010001
24. Lutz, M. P., & Zimmerman, R. W. (2021). The effect of pore shape on the Poisson ratio of porous
materials. Mathematics and Mechanics of Solids, 26(8), 1191-1203.
25. Madhiarasan, M., & Louzazni, M. (2022). Analysis of artificial neural network: architecture, types, and
forecasting applications. Journal of Electrical and Computer Engineering, 2022(1), 5416722.
https://doi.org/10.1155/2022/5416722
26. Mahdi, D. S., & Alrazzaq, A. A. (2023). Rock mechanical properties: A review of experimental tests
and prediction approaches. Iraqi Journal of Oil and Gas Research (IJOGR), 3(1), 106-115.
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue IX September 2025
Page 2749
27. Małkowski, P., Niedbalski, Z., & Balarabe, T. (2021). A statistical analysis of geomechanical data and
its effect on rock mass numerical modeling: a case study. International Journal of Coal Science and
Technology, 8(2), 312-323.
28. Millán, M. A., Galindo, R., & Alencar, A. (2021). Application of artificial neural networks for
predicting the bearing capacity of shallow foundations on rock masses. Rock Mechanics and Rock
Engineering, 54(9), 5071-5094.
29. Montesinos López, O. A., Montesinos López, A., & Crossa, J. (2022). Fundamentals of artificial neural
networks and deep learning. In Multivariate statistical machine learning methods for genomic
prediction (pp. 379-425). Cham: Springer International Publishing.
30. Obasohan, J. N., Igbokwe, J. I., & Ojanikele, W. A. (2021). Mapping and Analysis of Coastal Erosion
Vulnerability along Bonny Island Coastline, Rivers State Using Satellite Remote Sensing. International
Journal of Engineering Science and Computing.
31. Ogoro, M. (2014). Spatio–Temporal Changes in the Geomorphic Shoreline of Bonny Island. Journal of
Research in Humanities and Social Science, 2(11), 75-80.
32. Qiang, Z., Yasin, Q., Golsanami, N., & Du, Q. (2020). Prediction of reservoir quality from log-core and
seismic inversion analysis with an artificial neural network: A case study from the Sawan Gas Field,
Pakistan. Energies, 13(2), 486.
33. Saikia, P., Baruah, R. D., Singh, S. K., & Chaudhuri, P. K. (2020). Artificial Neural Networks in the
domain of reservoir characterization: A review from shallow to deep models. Computers &
Geosciences, 135, 104357. https://doi.org/10.1016/j.cageo.2019.104357
34. Salman, S., & Liu, X. (2019). Overfitting mechanism and avoidance in deep neural networks. arXiv
preprint arXiv:1901.06566. https://doi.org/10.48550/arXiv.1901.06566
35. Samek, W., Montavon, G., Lapuschkin, S., Anders, C. J., & Müller, K. R. (2021). Explaining deep
neural networks and beyond: A review of methods and applications. Proceedings of the IEEE, 109(3),
247-278.
36. Sewak, M., Sahay, S. K., & Rathore, H. (2020). An overview of deep learning architecture of deep
neural networks and autoencoders. Journal of Computational and Theoretical Nanoscience, 17(1), 182-
188.
37. Smys, S., Chen, J. I. Z., & Shakya, S. (2020). Survey on neural network architectures with deep
learning. Journal of Soft Computing Paradigm (JSCP), 2(03), 186-194.
38. Yan, B., Wang, P., Ren, F., Guo, Q., & Cai, M. (2020). A review of mechanical properties and
constitutive theory of rock mass anisotropy. Arabian journal of geosciences, 13, 1-16.
39. Yang, X., Shih, S. M., Fu, Y., Zhao, X., & Ji, S. (2022). Your vit is secretly a hybrid discriminative-
generative diffusion model. arXiv preprint arXiv:2208.07791.
https://doi.org/10.48550/arXiv.2208.07791
40. Zoorabadi, M., & Muralha, J. (2025). Shear Strength of Rock Discontinuities with Emphasis on the
Basic Friction Angle Based on a Compiled Database. Geotechnics, 5(3), 48.