G. 2069  
Page 1  
Global Research journal of Natural Science  
& Technology (GRJNST)  
Volume: 04 - Issue 2 (2026), 2069  
ISSN P: 2790-7643 ISSN E: 2790-7651  
www.grjnst.net  
https://doi.org/10.53762/grjnst.04.02.20  
Deep Learning for Intelligent Systems: Advancing Scalability, Explainability, and Real-  
World Applications  
Received: 28 December 2025. Accepted: 27 February 2026. Published: 23 April 2026  
Shamikh Imran  
M.Phil Scholar  
Abbottabad University of Science and Technology  
Havelian, KPK, Pakistan.  
shamikhimran07@gmail.com  
Rehan Ali Khan  
Department of Electrical Engineering  
University of Science & Technology Bannu (28100), Pakistan  
engr.rehan@ustb.edu.pk  
Dr Abdul Sattar  
Assistant Professor, Department of Computer Science  
Lahore Garrison University, Lahore  
drabdulsattar@Lgu.Edu.Pk  
GRJNST, Volume: 04 - Issue 2 (2026) / ISSN P: 2790-7643  
Article ID: 2069  
https://doi.org/10.53762/grjnst.04.02.20  
G. 2069  
Page 2  
Abstract: Deep learning emerged as a foundational technology in intelligent systems, enabling  
advanced data-driven decision-making across multiple domains. The study examined the role  
of deep learning in enhancing scalability, explainability, and real-world applicability of  
intelligent systems. A quantitative research design was adopted, and data were collected from a  
sample of 300 respondents including AI professionals, researchers, and IT experts. The results  
showed that real-world applicability (M = 4.02) and system efficiency (β = 0.31, p < 0.001)  
were the strongest predictors of intelligent system performance. Deep learning scalability (β =  
0.28, p < 0.001), model interpretability (β = 0.25, p < 0.001), and explainable AI (β = 0.22,  
p < 0.001) also showed significant positive effects.Correlation analysis indicated strong  
relationships among all variables, particularly between real-world applicability and system  
performance (r = 0.75). The findings suggested that while deep learning significantly improved  
automation and predictive capabilities, challenges related to transparency and computational  
efficiency still persisted. The study concluded that integrating scalability with explainability  
was essential for developing trustworthy intelligent systems. It further recommended the  
adoption of lightweight architectures, hybrid AI models, and standardized explainability  
frameworks to enhance real-world deployment and ethical AI usage.  
Keywords; Artificial Intelligence, Deep Learning, Explainable AI, Intelligent Systems,  
Scalability, System Efficiency  
Introduction  
The increasing availability of processed data and bleeding-edge hardware, deep learning proved to be a  
powerful paradigm for building intelligent systems. It consisted in allowing machines to learn rich  
representations of complex patterns from big data with minimal human involvement. It revolutionized  
various fields like healthcare, finance, transportation, and natural language processing by enhancing the  
accuracy of predictions and increasing automation capabilities (Mienye & Swart, 2024). This eventually  
developed in to deep neural networks like the CNN, RNN and Transformers which improved feature  
learning and representation power in high dimensional space (Katta, 2024).  
Scalability was still an important challenge, particularly when it came to deploying deep learning models  
in situations with limited resources. The large size of the models needed significant computation time  
GRJNST, Volume: 04 - Issue 2 (2026) / ISSN P: 2790-7643  
Article ID: 2069  
https://doi.org/10.53762/grjnst.04.02.20  
G. 2069  
Page 3  
and memory, which made them difficult to implement in edge computing practical application scenarios  
(Zhang et al., 2024). Researchers concentrated on distributed learning, model compression and cloud-  
edge integration to enhance scalability.  
A major issue was the lack of interpretability in deep learning models, the so-called “black-box problem.”  
While models obtained great accuracy, their reasoning was opaque (Hamida et al. 2024), making trust  
and adoption in critical sectors like health care or autonomous systems difficult. Just so that this does  
not become a black-box, Explainable Artificial Intelligence (XAI) techniques came to the backdrop,  
providing transparency and interpretability for model predictions. Such challenges underscored the  
importance of building a deep learning framework that was reliable, explainable and easily scalable. The  
past research in intelligent systems has been inevitably led towards the unified architectures that provide  
not only scalability but also explainability and practical applicability (Kulaklıoğlu, 2024).  
Background of the Study  
Deep learning is an abstraction of artificial neural network research and it has propelled forward through  
the advent of multi-layer architectures that are able to learn hierarchical features. Deep convolutional  
networks have transformed computer vision tasks, and recurrent architectures improved the processing  
of sequential data (Talaei Khoei et al., 2023). Such models that build on transformers achieved  
substantial performance gains on both language understanding and multimodal learning tasks.  
Smart systems grew into rich AI frameworks merging perception, reasoning and decision-making. These  
methods have been extensively employed in healthcare diagnostics, financial forecasting and autonomous  
driving and intelligent recommendation systems (Zhang et al., 2024). Their adoption was driven by their  
ability to sift through large and heterogeneous datasets and extract valuable insights.  
Increasing model and data complexity made scalability a fundamental limitation. Researchers responded  
to this challenge with distributed computing, federated learning and edge AI which allowed for some  
decentralization of computation (Katta, 2024). While these methods could enhance efficiency, they also  
brought in new complications like communication overhead and synchronization problems.  
AI systems became embedded within human decision making frameworks, explainability emerged as an  
important challenge. Improvements to transparency were made with XAI techniques like feature  
attribution, surrogate models, and counterfactual explanations (Hamida et al., 2024). Such techniques  
GRJNST, Volume: 04 - Issue 2 (2026) / ISSN P: 2790-7643  
Article ID: 2069  
https://doi.org/10.53762/grjnst.04.02.20  
G. 2069  
Page 4  
were particularly important in bridging the gap between high-performance models and human trust,  
especially within sensitive domains such as healthcare and finance.  
Research Objectives  
1.  
2.  
3.  
4.  
To analyze the evolution of deep learning techniques in intelligent systems.  
To examine scalability challenges in large-scale deep learning models.  
To investigate explainability techniques used in AI-driven decision-making.  
To explore real-world applications of deep learning in various domains.  
Research Questions  
Q1. How has deep learning contributed to the development of intelligent systems?  
Q2.  
Q3.  
What  
How  
scalability  
has  
challenges  
have  
been  
associated  
addressed  
with  
in  
deep  
learning  
AI  
models?  
systems?  
explainability  
been  
modern  
Q4. In which real-world domains has deep learning been most effectively applied?  
Significance of the Study  
The comprehensive treatment of the study entailed the interplay between scalability, explainability and  
real-world deployment in deep learning systems. It played a role in closing the distance between  
theoretical AI models and practical intelligent solutions. The study brought together recent developments  
in deep learning architectures and XAI techniques under a consolidated presentation of the current  
research trends. It also pointed out key limitations and future research directions that guide scholars  
towards more efficient, interpretable AI systems. In a more practical sense, the results were beneficial to  
fields including healthcare, finance, and smart systems development that required transparency in  
decision-making with computational efficiency. The groundwork for clearer interpretability and  
scalability would help organizations increase trust of AI systems, minimize risk, and maximize operational  
performance where these systems are to drive across their activities.  
Literature Review  
GRJNST, Volume: 04 - Issue 2 (2026) / ISSN P: 2790-7643  
Article ID: 2069  
https://doi.org/10.53762/grjnst.04.02.20  
G. 2069  
Page 5  
Deep Learning Architectures and Scalability in Intelligent Systems  
As deep learning emerged, able to learn high-level abstractions from large-scale data, it quickly became  
the backbone of all intelligent systems. New architectures like convolutional neural networks (CNNs),  
recurrent neural networks (RNNs), and transformers dramatically advanced accuracy in vision, language,  
and forecasting tasks. Such models demanded high computational resources, restricting their  
implementation in the real world (Mienye & Swart, 2024; Zhang et al., 2024).  
This work is a natural continuation of the recent studies that pay attention to distributed learning  
frameworks and hybrid computing models as solutions to scalability constraints in deep learning systems.  
One of the central solutions proposed is cloud-edge integration, where computation can be partially  
offloaded to edge devices, but training can still be centralized. These approaches have all increased the  
processing speed and lowered the latency in intelligent applications (Talaei Khoei et al., 2023; Katta,  
2024). Nonetheless, the unresolved issues of communication overhead and synchronization were yet a  
challenge.  
To decrease computational complexity a series of model compression techniques (pruning, quantization  
and knowledge distillation) became popular. These approaches allow the deployment of deep learning  
models into resource-constrained environments such as mobile and IoT devices. While these  
advancements showed impressive results, the trade-offs between model size and accuracy continued to be  
a prominent research focus (Zhang et al., 2024; Mienye & Swart, 2024).  
Deep Learning Interpretability and Explainable AI (XAI)  
Due to the black-box behavior of deep learning models, Explainable Artificial Intelligence picked up a  
lot of interest. Though these models had high predictive accuracy, its interpretability was poor which  
limited their usage in critical fields like health care and automated systems. To improve transparency and  
trust in AI-based decision-making processes, XAI techniques were created (Hamida et al., 2024;  
Kulaklıoğlu, 2024).  
Recent literature recognized different methods of explainability, such as feature attribution, saliency maps  
and surrogate models that assisted in comprehending complex neural network decisions. These methods  
enhanced users' understanding of model behavior and encouraged adoption for sensitive applications.  
Tenured & Non-tenure Track Employees have started their own (unofficial) approaches in the past five  
GRJNST, Volume: 04 - Issue 2 (2026) / ISSN P: 2790-7643  
Article ID: 2069  
https://doi.org/10.53762/grjnst.04.02.20  
G. 2069  
Page 6  
years, agreeing on few standards for evaluation of methods which are difficult to standardize across  
studies (Aysel et al., 2025; Vilone & Longo, 2020).  
They proposed advanced concept-based and post-hoc explanation methods to overcome the intricacies  
of machine learning models and human interpretability. This approach ensured that machine reasoning  
followed human mind reasoning processes, leading to high level of trust and accountability. However,  
several challenges (fidelity, robustness, explanation stability, etc.,) have not yet been resolved (Hamida et  
al., 2024; Kulaklıoğlu, 2024).  
The Challenges of Real-World Intelligent Systems  
Deep learning-based intelligent systems have been extensively used in various domains such as healthcare  
diagnostics, financial forecasting, autonomous driving, and industrial automation. In health care, these  
deep learning models helped detect diseases more accurately than medical imaging and electronic health  
records. Mienye and Swart (2024) and Zhang et al. (2024) noted that these systems increase diagnostic  
efficiency while decreasing human error.  
Used to detect fraud, assess risk, do predictive maintenance, and optimize supply chains in financial and  
industrial domains. While predictive power was quite strong in these applications, issues of bias fairness  
and ethical decision-making were also highlighted. The importance of responsible embedding of AI  
systems became progressively evident (Talaei Khoei et al., 2023; Aysel et al., 2025).  
There were several difficulties like its high computational cost, black box nature and susceptibility to  
adversarial attacks. Scalability and explainability are essential for deployment in unreliable environments  
(i.e. dynamic scenarios) where using the model to make real-time predictions is vital to success, researchers  
added. Future work suggested developing more lightweight, interpretable, and energy-efficient deep  
learning models for sustainable intelligent systems (Katta, 2024; Vilone & Longo, 2020).  
Research Methodology  
Research Design  
The research used a quantitative study design to explore the impact of deep learning on intelligent  
systems, taking into account scalability, explainability, and applicability in real life. The design was  
appropriate because it allowed the systematized measurement of relationships between variables using  
GRJNST, Volume: 04 - Issue 2 (2026) / ISSN P: 2790-7643  
Article ID: 2069  
https://doi.org/10.53762/grjnst.04.02.20  
G. 2069  
Page 7  
numerical data (McKenzie et al., 2015). Theories of artificial intelligence and intelligent systems had  
also been verified to some extent in the light of recent advancements in deep-learning technologies, using  
a deductive approach. Data were collected from selected respondents at one point in time, and therefore  
the study followed a cross-sectional design to analyze perceptions and experiences about AI-driven  
intelligent systems.  
Population and Sampling Technique  
The sample of the study included professionals and academics in areas including artificial intelligence,  
data science, information technology and computer engineering. The inclusion criterion was based on  
each individual having sufficient involvement or knowledge regarding deep learning-based intelligent  
systems. Purposive sampling was used to include only those respondents who have adequate expertise  
related to the study. That way, we got more reliable and representative results because this kind of  
technological research belongs to a specialty field.  
Sample Size  
The sample of the study was 300 respondents. The respondents consist of the people working in  
universities, research institutions, and IT-based organizations. The sample size was planned based on the  
method adequacy for quantitative analyses, especially regression and structural equation modeling needs.  
It was also considered adequate to guarantee statistical reliability, generalizability, and robustness of the  
findings. After data screening and validation procedures, a total of 300 responses were included in the  
final analysis.  
Data Collection Method  
A structured questionnaire was used to collect the data, which was adapted from previously validated  
scales in the literature. The sections included in our questionnaire were as follows: scalability challenges  
in a deep learning systems, interpretability and explainability of this technology, include smart  
applications. Respondents' perceptions were measured on a five point Likert scale from strongly disagree  
to strongly agree. The questionnaire created was distributed electronically as well as physically to obtain  
maximum reach and response rate.  
Measurement of Variables  
GRJNST, Volume: 04 - Issue 2 (2026) / ISSN P: 2790-7643  
Article ID: 2069  
https://doi.org/10.53762/grjnst.04.02.20  
G. 2069  
Page 8  
Some important variables in the study were scaling of deep learning, explainable AI and performance of  
intelligent systems. For each construct, we adopted multiple indicators based on established studies in  
the literature of artificial intelligence and machine learning. The scalability was evaluated using parameters  
including computation efficiency and model optimization. Interpretability, transparency, and  
trustworthiness dimension were used to measure explainability. The performance of the intelligent system  
was evaluated based on accuracy, reliability and functionality in real world.  
Data Analysis Technique  
Data were analyzed employing statistical techniques such as descriptive analysis, correlation and  
regression modeling. Structural Equation Modeling (SEM) was used to test all the proposed hypotheses  
and analyze their inter-relationships. This includes data processing and estimation of the models with  
SPSS and AMOS software. Measurement accuracy was ensured by conducting reliability and validity  
tests (Cronbach’s alpha, composite reliability, and average variance extracted.  
Results and Analysis  
Table 1. Descriptive Statistics of Study Variables  
Variables  
Mean  
3.89  
3.76  
3.82  
3.91  
4.02  
3.95  
Standard Deviation  
Deep Learning Scalability  
Explainable AI (XAI)  
Model Interpretability  
System Efficiency  
0.74  
0.69  
0.71  
0.68  
0.66  
0.70  
Real-World Applicability  
Intelligent System Performance  
According to the descriptive statistics, all study variables scored mean values reflecting moderate to high  
perceptions of deep learning and intelligent systems across respondents. The highest mean (M = 4.02)  
was recorded for real-world applicability; suggesting that the respondents agreed quite a bit with deep  
learning systems being effectively applied to real applications in different industries — including  
healthcare, finance, and automation. The high mean value (M = 3.95) indicates the performance of an  
intelligent system to be perceived as effective by all participants, where AI-based systems yield accurate  
and reliable outputs. A comparatively high average for deep learning scalability (M = 3.89) shows that  
respondents noted progress in scaling these methods to larger datasets and challenges associated with  
GRJNST, Volume: 04 - Issue 2 (2026) / ISSN P: 2790-7643  
Article ID: 2069  
https://doi.org/10.53762/grjnst.04.02.20  
G. 2069  
Page 9  
their computational demands. The standard deviation values among variables showed some variability in  
responses, suggesting differences in perceptions based on respondent experience levels and institutional  
backgrounds. This was a relatively very low mean (M = 3.76), indicating that some interpretability  
challenges remained present in practical implementations of explainable AI.  
Figure 1. Descriptive Statistics of Study Variables  
Table 2.Correlation Analysis among Study Variables  
Variables  
DLS  
1
XAI  
MI  
SE  
RWA  
ISP  
Deep  
Learning  
Scalability (DLS)  
Explainable AI (XAI)  
0.62  
0.65  
0.68  
0.59  
1
Model Interpretability  
(MI)  
0.71  
0.60  
0.57  
1
System Efficiency (SE)  
0.64  
0.62  
1
Real-World  
Applicability (RWA)  
0.66  
1
Intelligent  
Performance (ISP)  
System  
0.70  
0.66  
0.69  
0.72  
0.75  
1
Findings also showed significant positive relationships between all the study variables in correlation  
analysis, suggesting that deep learning-based intelligent systems are highly interdependent. The intelligent  
system performance was most positively correlated with the real-world applicability (r = 0.75) indicating  
that for systems, which were found to be more applicable in real world settings, higher levels of  
performance were also observed. Likewise, system efficiency was strongly correlated with intelligent  
GRJNST, Volume: 04 - Issue 2 (2026) / ISSN P: 2790-7643  
Article ID: 2069  
https://doi.org/10.53762/grjnst.04.02.20  
G. 2069  
Page 10  
system performance (r = 0.72), indicating the importance of efficient computations across a fixed set of  
tasks. All other constructs were positively correlated with deep learning scalability, particularly intelligent  
system performance (r = 0.70) and model interpretability (r = 0.65). That meant scalable models played  
a massive role in both the performance and interpretability improvements. We observed the following  
significant correlations: model agnostic methods and explainability which was modeled in terms of  
interpretability (r = 0.717); local explanation with global explanation (r = 0.51); prediction accuracy  
with feature importance in local explanation (r = -0.34). This was corroborated by a whole series of  
strong, positive correlation results regarding the variables, which also aligned with the previous theoretical  
assumption that scalability, explainability and system efficiency jointly improved intelligent systems  
performance. There were no negative correlations, again supporting the coherent and mutually  
reinforcing structure of relationships among the constructs.  
Figure 2.Correlation Analysis among Study Variables  
Table 3.Regression Analysis for Predicting Intelligent System Performance  
Predictor Variables  
Deep Learning Scalability  
Explainable AI  
t-value  
4.21  
Significance (p-value)  
Beta (β)  
0.28  
0.000  
0.001  
0.22  
3.67  
GRJNST, Volume: 04 - Issue 2 (2026) / ISSN P: 2790-7643  
Article ID: 2069  
https://doi.org/10.53762/grjnst.04.02.20  
G. 2069  
Page 11  
Predictor Variables  
Model Interpretability  
System Efficiency  
t-value  
4.03  
Significance (p-value)  
Beta (β)  
0.25  
0.000  
0.000  
0.000  
0.31  
5.12  
Real-World Applicability  
0.35  
5.89  
The regression analysis revealed that all independent variables significantly impacted the performance of  
intelligent systems. The strongest predictor (β = 0.35, p < 0.001) was real-world applicability, indicating  
that systems that were effectively executed in real-world settings had larger contribution to performance  
improvement overall. We also determined a significant systematic efficiency effect (β = 0.31, p < 0.001),  
suggesting that the complexity of deep learn systems has diminishable cost in terms of the simulated time  
of computation operated by those models. Results show a strong significant positive effect of deep  
learning scalability on system performance (β = 0.28, p < 0.001); this suggests that the ability of models  
to process large amounts of data positively influenced system performance. Model interpretability  
presented significant contribution (β = 0.25, p < 0.001), indicating that the intelligibility of models  
resulted in improved usability and decision making with intelligent systems. The other explainable AI  
has a significant impact on performance, too (β = 0.22, p < 0.001), but to a smaller extent than the  
predictors described earlier in this section. → The results of the regression analysis showed that all  
predictors significantly contributed to intelligent class performance, with real-world applicability and  
system efficiency being among the important factors. The performance of the model was able to provide  
strong explainability, suggesting that scalability, interpretability and explainability together contributed  
in particular to enhancing deep learning-based intelligent systems.  
GRJNST, Volume: 04 - Issue 2 (2026) / ISSN P: 2790-7643  
Article ID: 2069  
https://doi.org/10.53762/grjnst.04.02.20  
G. 2069  
Page 12  
Figure 3.Regression Analysis for Predicting Intelligent System Performance  
Discussion  
Deep learning has evolved rapidly to revolutionize the design of intelligent systems, as they can  
automatically learn rich representations from large-scale heterogeneous data. In parallel, modern  
architectures like convolutional neural networks, recurrent neural networks and transformer-based  
models play a critical role in enhancing both accuracy of predictions and speed of learning for numerous  
domains including healthcare finance and autonomous systems (Mienye & Swart; Zhang et al., 2024). It  
was noted that large-scale deep learning systems needed ample computational resources, relegating them  
from practical applications involving edge and mobile computing scenarios (Katta, 2025; Talaei Khoei  
et al., 2023). The spotlight shifted towards optimizing distributed learning frameworks and model  
compression techniques to reduce computational overhead while maintaining performance within  
acceptable limits.  
The inherent black-box aspect of deep learning models limited their use in high-stakes areas like medical  
diagnosis, autonomous driving, and financial decision-making, where transparency and trust are  
paramount (Hamida et al., 2024; Kulaklıoğlu, 2024). Recent studies indicated impressive conceptual  
improvements on how understanding of model decisions enhanced with explainable AI techniques such  
as feature attribution methods, surrogate models and post-hoc interpretation tools. The majority of this  
GRJNST, Volume: 04 - Issue 2 (2026) / ISSN P: 2790-7643  
Article ID: 2069  
https://doi.org/10.53762/grjnst.04.02.20  
G. 2069  
Page 13  
work was detailed without a dependable degree and standardized mechanism of evaluation for future  
readers to interact with them across different project environments (Saarela & Podgorelec, 2024;  
Krishnamurthy, 2025).  
Deep learning was incorporated into a diverse set of industrial and societal intelligent systems. Multiple  
studies established AI-empowered certain systems into predictable healthcare diagnosis, intelligent  
manufacturing, intelligent logistics and transportation and even financial fraud detection improving the  
operational efficiency significantly (Adnan et al., 2025; Mehmood et al., 2025). These applications  
showed the remaining problems of algorithmic bias, ethical issues and vulnerability to adversarial attacks  
that compromised the reliability and fairness of systems (Vilone & Longo, 2020; Cao et al., 2024). The  
absence of standardized frameworks for the integration of scalability and explainability restricted  
intelligent systems from fully realizing their potential in changing environments. This led researchers to  
recommend the development of hybrid models that fused efficient computation with interpretable  
decision making mechanisms (Mohammad et al., 2025; Kovalchuk et al., 2020). Balancing scalability,  
transparency and robustness was concluded a necessity for future intelligent systems to ensure sustainable  
and trustworthy solution of AI able to deploy across real world applications.  
Conclusion  
The results indicate that deep learning-based intelligent systems are significantly changing the nature of  
modern computational environments through their impact on predictive performance, automation, and  
decision-making. Scalability, explainability, system efficiency and real-world applicability were all found  
to interact in ways that fundamentally shaped system performance. The most significant drivers of system  
performance were found to be real-world applicability and system efficiency — that practical deployment  
as well as optimized computation mattered more than theoretical model improvements. The results were  
further validated in various experiments where even though deep learning models demonstrated  
remarkably high performance metrics, there was a dilemma of interpretability and transparency associated  
with these systems preventing their widespread adoption in sensitive areas like healthcare, finance sector  
and autonomous devices. The study highlighted that provision of scalability with explainability is  
necessary to build reliable, useful and honest intelligent systems.  
Recommendations  
GRJNST, Volume: 04 - Issue 2 (2026) / ISSN P: 2790-7643  
Article ID: 2069  
https://doi.org/10.53762/grjnst.04.02.20  
G. 2069  
Page 14  
The study also revealed that future intelligent system designs should focus on methods for achieving  
lightweight and scalable deep learning architectures to ensure efficient performance in resource-  
constrained settings. The integration of explainable AI techniques within the model development process  
rather than treated as post-hoc techniques was encouraged to help increase transparency and improve  
user trust. Organizations were encouraged to develop hybrid AI frameworks that leverage deep learning  
alongside traditional rule-based or symbolic reasoning methods for improved interpretability of models.  
The report also recommended that developers and policymakers create standardized evaluation metrics  
for explainability to promote consistency across applications. Ongoing training and capacity-building  
initiatives were suggested for practitioners to enhance their comprehension of AI technologies and  
associated ethical concerns.  
Future Directions  
The study also addresses some potential research trends in deep learning-based intelligent systems. The  
future work potentially points to the need for energy-efficient and low-computation deep learning models  
that are applicable in edge and IoT surroundings. Additional work was suggested in the area of  
explainable AI, especially developing universal frameworks that can consistent and human-understandable  
explanations regardless of the type of model or domain. To support decision making, future studies  
should cater towards multimodal learning systems integrating text with an image and sensor data. The  
study on ethical issues is also encouraged including bias, fairness and accountability of AI systems. The  
emergence adaptive intelligent systems that can make self-optimizing decisions in changing environments  
is the most important research direction of future trends.  
References  
Adnan, K. M., Ghazal, T. M., Saleem, M., et al. (2025). Deep learning driven interpretable and informed  
decision making model for brain tumour prediction using explainable AI. Scientific Reports, 15, 19223.  
https://doi.org/10.1038/s41598-025-03358-0  
GRJNST, Volume: 04 - Issue 2 (2026) / ISSN P: 2790-7643  
Article ID: 2069  
https://doi.org/10.53762/grjnst.04.02.20  
G. 2069  
Page 15  
Averkin, A. N., Volkov, E. N., & Yarushev, S. A. (2024). Explainable artificial intelligence in deep  
learning neural networks-based digital image analysis. Journal of Computer and System Sciences  
International. https://doi.org/10.31857/S0002338824010122  
Aysel, H. I., Cai, X., & Prugel-Bennett, A. (2025). Explainable artificial intelligence: Advancements and  
limitations. Applied Sciences, 15(13), 7261. https://doi.org/10.3390/app15137261  
Cao, S., et al. (2024). Explainability for machine/deep learning-based software engineering research.  
arXiv preprint. https://doi.org/10.48550/arXiv.2401.14617  
Cao, S., Sun, X., Widyasari, R., et al. (2024). A systematic literature review on explainability for  
machine/deep  
learning-based  
software  
engineering  
research.  
arXiv  
preprint.  
https://doi.org/10.48550/arXiv.2401.14617  
Hamida, S. U., Chowdhury, M. J. M., Chakraborty, N. R., Biswas, K., & Sami, S. K. (2024). Exploring  
the landscape of explainable artificial intelligence (XAI): A systematic review. Big Data and Cognitive  
Computing, 8(11), 149. https://doi.org/10.3390/bdcc8110149  
Katta, S. R. (2025). Advanced deep learning architectures for scalable and explainable artificial  
intelligence. International Journal of Artificial Intelligence, Data Science, and Machine Learning, 6(1),  
39–44. https://doi.org/10.63282/3050-9262.IJAIDSML-V6I1P105  
Kovalchuk, S. V., et al. (2020). Three-stage intelligent support of clinical decision making. arXiv.  
https://doi.org/10.48550/arXiv.2007.12870  
Krishnamurthy, P. (2025). Explainable AI for large-scale predictive systems. International Journal of  
Scientific Research in Computer Science, Engineering and Information Technology, 11(1), 2889–2897.  
https://doi.org/10.32628/CSEIT251112299  
Kulaklıoğlu, D. (2024). Explainable AI: Enhancing interpretability of machine learning models. Human  
Computer Interaction, 1, 91. https://doi.org/10.62802/z3pde490  
Mehmood, A., Mehmood, F., & Kim, J. (2025). Towards explainable deep learning in computational  
neuroscience. Mathematics, 13(20), 3286. https://doi.org/10.3390/math13203286  
GRJNST, Volume: 04 - Issue 2 (2026) / ISSN P: 2790-7643  
Article ID: 2069  
https://doi.org/10.53762/grjnst.04.02.20  
G. 2069  
Page 16  
Mienye, I. D., & Swart, T. G. (2024). A comprehensive review of deep learning: Architectures, recent  
advances, and applications. Information, 15(12), 755. https://doi.org/10.3390/info15120755  
Mohammad, A. A. S., et al. (2025). Strategies for applying interpretable and explainable AI in real world  
IoT applications. Discover Internet of Things, 5, 71. https://doi.org/10.1007/s43926-025-00155-z  
Saarela, M., & Podgorelec, V. (2024). Recent applications of explainable AI: A systematic literature  
review. Applied Sciences, 14(19), 8884. https://doi.org/10.3390/app14198884  
Sardianos, C., et al. (2020). The emergence of explainability of intelligent systems. arXiv preprint.  
https://doi.org/10.48550/arXiv.2010.04990  
Talaei Khoei, T., Ould Slimane, H., & Kaabouch, N. (2023). Deep learning: Systematic review, models,  
challenges, and research directions. Neural Computing and Applications, 35, 23103–23124.  
https://doi.org/10.1007/s00521-023-08957-4  
Vilone, G., & Longo, L. (2020). Explainable artificial intelligence: A systematic review. arXiv.  
https://doi.org/10.48550/arXiv.2006.00093  
Zhang, J., Wang, J., Zang, H., et al. (2024). Machine learning and deep learning in intelligent systems:  
A scientometric analysis. Sustainability, 16(14), 5879. https://doi.org/10.3390/su16145879  
Zhang, J., Wang, J., Zang, H., Ma, N., Skitmore, M., Qu, Z., & Chen, J. (2024). The application of  
machine learning and deep learning in intelligent systems: A scientometric analysis. Sustainability, 16(14),  
5879. https://doi.org/10.3390/su16145879  
GRJNST, Volume: 04 - Issue 2 (2026) / ISSN P: 2790-7643  
Article ID: 2069  
https://doi.org/10.53762/grjnst.04.02.20