Authors

  • Oleksii Segeda
    Senior Data Engineer, Mapbox Washington, D.C., USA

DOI:

https://doi.org/10.37547/tajet/Volume07Issue05-08

Keywords:

geospatial intelligence machine learning intelligent search deep learning data integration GEOINT anomaly detection semantic segmentation

Abstract

This article explores the potential for improving intelligent search through the integration of geospatial data and machine learning techniques. It reviews current approaches in the field of GEOINT, including the processing of satellite imagery, vector data, and crowd-sourced sources such as OpenStreetMap, along with the application of deep learning architectures (e.g., VGG16, U-Net) and anomaly detection algorithms (e.g., Isolation Forest, One-Class SVM). A comprehensive literature review is provided, highlighting the relevance of the topic and identifying a research gap stemming from the lack of a holistic interdisciplinary framework. In response, the article proposes an integrated methodology aimed at increasing the accuracy and interpretability of intelligent search systems. Based on empirical data derived from modern computational platforms and multimodal models, the study demonstrates that combining geospatial data with intelligent search algorithms opens new opportunities for building adaptive and high-precision analytical systems capable of responding quickly to dynamic environmental changes. The findings are of interest to professionals and researchers in geoinformatics and machine learning seeking to merge analytical methods to improve the performance of intelligent search systems with spatial data. Additionally, the approaches discussed may prove valuable in interdisciplinary research related to decision-making optimization in fields such as urban planning, logistics, and environmental monitoring.


background image

The American Journal of Engineering and Technology

101

https://www.theamericanjournals.com/index.php/tajet

TYPE

Original Research

PAGE NO.

101-108

DOI

10.37547/tajet/Volume07Issue05-08



OPEN ACCESS

SUBMITED

24 March 2025

ACCEPTED

20 April 2025

PUBLISHED

12 May 2025

VOLUME

Vol.07 Issue 05 2025

CITATION

Oleksii Segeda. (2025). Enhancing Search Intelligence with Geospatial Data
and Machine Learning. The American Journal of Engineering and
Technology, 7(05), 101

108.

https://doi.org/10.37547/tajet/Volume07Issue05-08.

COPYRIGHT

© 2025 Original content from this work may be used under the terms
of the creative commons attributes 4.0 License.

Enhancing Search
Intelligence with
Geospatial Data and
Machine Learning

Oleksii Segeda

Senior Data Engineer, Mapbox Washington, D.C., USA

Abstract:

This article explores the potential for

improving intelligent search through the integration of
geospatial data and machine learning techniques. It
reviews current approaches in the field of GEOINT,
including the processing of satellite imagery, vector
data,

and

crowd-sourced

sources

such

as

OpenStreetMap, along with the application of deep
learning architectures (e.g., VGG16, U-Net) and
anomaly detection algorithms (e.g., Isolation Forest,
One-Class SVM). A comprehensive literature review is
provided, highlighting the relevance of the topic and
identifying a research gap stemming from the lack of a
holistic interdisciplinary framework. In response, the
article proposes an integrated methodology aimed at
increasing the accuracy and interpretability of
intelligent search systems. Based on empirical data
derived from modern computational platforms and
multimodal models, the study demonstrates that
combining geospatial data with intelligent search
algorithms opens new opportunities for building
adaptive and high-precision analytical systems capable
of responding quickly to dynamic environmental
changes. The findings are of interest to professionals
and researchers in geoinformatics and machine
learning seeking to merge analytical methods to
improve the performance of intelligent search systems
with spatial data. Additionally, the approaches
discussed may prove valuable in interdisciplinary
research related to decision-making optimization in
fields such as urban planning, logistics, and
environmental monitoring.

Keywords:

geospatial intelligence, machine learning,

intelligent search, deep learning, data integration,


background image

The American Journal of Engineering and Technology

102

https://www.theamericanjournals.com/index.php/tajet

GEOINT, anomaly detection, semantic segmentation.

Introduction:

The integration of geospatial data and

machine learning methods represents a promising
direction for improving information retrieval, decision
support, and resource management in fields such as
national security, environmental monitoring, and urban
planning [2]. The use of high-resolution satellite
imagery (e.g., Sentinel-2, Landsat) and open-source
platforms such as OpenStreetMap provides a rich
foundation for analysis. When combined with modern
machine learning algorithms, these data sources open
new possibilities for intelligent search systems [1].

The literature reveals several major research
directions, each making a substantial contribution to
the development of both geospatial analysis and
intelligent search technologies. The first group of
studies focuses on geospatial intelligence and data
management. For example, Kolluru V. et al. [1] present
a systematic review of current approaches for
improving geospatial intelligence using advanced data
analytics and machine learning. Their work
demonstrates how the use of large volumes of
heterogeneous data can significantly enhance spatial
analysis outcomes. In a similar vein, Breunig M. et al. [3]
explore the evolution of geodata management,
highlighting major achievements and identifying future
development opportunities in the face of emerging
challenges. A practical perspective is reflected in the
work of Feldmeyer D. et al. [6], who use
OpenStreetMap data and machine learning to generate
socio-economic

indicators

an

example

of

interdisciplinary implementation. Gromny E. et al. [7]
further contribute by developing a training dataset for
land cover classification using Sentinel-2 imagery,
which significantly improves the quality and accuracy of
geospatial analysis.

A second group of publications centers on the use of
machine learning to enhance the performance of
intelligent search and information retrieval. Ghadge N.
[2] focuses on optimizing search processes, showing
how machine learning algorithms can improve the
relevance and accuracy of retrieved information.
Similarly, Kolluru V., Mungara S., and Chintakunta A. N.
[4] introduce tools for combating misinformation using
machine learning to filter unreliable data and
strengthen the reliability of news sources. In this

context, Bhatt S. et al. [10] emphasize semantic
enrichment of input data using knowledge graphs,
significantly improving query interpretation in AI
systems and thereby enhancing their effectiveness.

Another line of research explores machine learning
applications in specific domains. Mungara S., Koganti S.,
Chintakunta A. N., Kolluru V. K., and Nuthakki Y. [5]
analyze consumer behavior in e-commerce, using
analytical models to uncover hidden patterns
influencing market dynamics. Wang J. et al. [9] trace the
evolution of machine learning over the past three
decades, demonstrating how these methods have been
applied to optimize wireless networks, illustrating their
wide applicability beyond pure information retrieval
tasks.

Equally important is the growing field of explainable AI.
Páez A. [8] calls for a pragmatic shift toward algorithmic
transparency, arguing that interpretability is essential
for integrating AI systems into critical information
infrastructure.

In summary, contemporary literature reveals a certain
tension between technical and methodological
approaches to integrating machine learning with
geospatial data. On one hand, the emphasis is on
merging diverse data sources and optimizing
algorithms for more accurate spatial and informational
analysis. On the other, there are methodological
discrepancies in how the effectiveness and real-world
applicability of these models are defined. Issues related
to data standardization, methodological consistency,
and ethical considerations remain insufficiently
addressed,

indicating

a

need

for

further

interdisciplinary research and the development of
comprehensive solutions.

The aim of this article is to analyze methods for
enhancing intelligent search by integrating geospatial
data and machine learning.

The scientific contribution lies in the synergistic
combination of deep learning techniques with
geospatial analysis to enable a comprehensive
approach to intelligent search. Unlike traditional
methods, the proposed approach not only improves
classification and segmentation accuracy but also
enhances

the

interpretability

of

results

by

incorporating spatial context. This interdisciplinary


background image

The American Journal of Engineering and Technology

103

https://www.theamericanjournals.com/index.php/tajet

methodology offers new prospects for solving critical
tasks in domains that require fast and accurate analysis
of large-scale data.

The author’s hypothesis is that integrating geospatial

data with modern machine learning algorithms can
significantly improve the accuracy and completeness of
information retrieval. It is assumed that combining
high-quality satellite data with efficient models for
classification, semantic segmentation, and anomaly
detection will lead to a deeper understanding of data
structures, thereby increasing query relevance and the
overall quality of extracted information.

The methodological framework of this study is based on
a review of recent research in geospatial intelligence
and machine learning, with a focus on their application
in intelligent search systems.

1. Geospatial Intelligence: Data Sources and
Contemporary Challenges

The early development of geospatial intelligence was
marked by manual collection, processing, and analysis
of cartographic data

a labor-intensive and error-

prone process. With the advent of satellite
technologies such as Landsat in the 1970s, and the
subsequent launch of programs like Sentinel-2,
analytical capabilities expanded dramatically, enabling
high-quality, near-real-time observation of land cover,
infrastructure changes, and environmental dynamics
[2]. Today, geospatial intelligence (GEOINT) relies not
only on high-resolution satellite imagery but also on
crowdsourced

GIS

platforms

most

notably,

OpenStreetMap. The integration of user-contributed
data from around the world allows for the creation of
detailed information models of terrain and
infrastructure, capturing even the smallest urban
features and landscape transformations [3].

Geospatial data, by nature, combine high spatial and
temporal granularity with the ability to unify
heterogeneous formats

raster imagery, vector

features, and time series. This integration equips
researchers with a broad analytical toolkit, from land
use monitoring to ecological modeling and territorial

management optimization.

Among the key methods for change detection are the
following:

Sentinel-2 (part of the Copernicus

program) offers multispectral optical imagery with
spatial resolution ranging from 10 to 60 meters. Its high
revisit frequency (about every 5 days using both
satellites) and wide spectral coverage

including near-

infrared and red-edge bands

enable:

vegetation monitoring using indices

such as NDVI and EVI;

timely detection of changes in

agricultural and natural ecosystems;

rapid response to emergencies (e.g.,

wildfires, floods, landslides) thanks to near-real-time
coverage of large areas.

Landsat, jointly operated by USGS and

NASA, provides one of the longest-standing archives of
Earth observation data, dating back to the early 1970s.
With resolution of ~30 meters in most spectral bands
and 15 meters in panchromatic mode, Landsat imagery
supports:

retrospective analysis of landscape

changes over decades;

identification of urban expansion,

agricultural

intensification,

and

ecosystem

degradation;

calibration

and

validation

of

contemporary remote sensing products using historical
datasets.

OpenStreetMap (OSM) is a global,

crowdsourced project maintained by a community of
volunteers. It offers vectorized geometries of
infrastructure

(streets,

buildings,

waterways),

transportation networks, and place names. OSM's main
advantage lies in its continuous updates and expansive
coverage [1, 3].


background image

The American Journal of Engineering and Technology

104

https://www.theamericanjournals.com/index.php/tajet

Table 1. Comparison of key sources of geospatial data [1

3]

Data Source

Description

Applications

Key

Advantages

Limitations

Sentinel-2

High-quality optical
satellite

imagery

with

multispectral

data (10

60 m)

Land

cover

monitoring,
agriculture,
emergency
response

High

revisit

rate, near-real-
time

access,

spectral
diversity

Limited geographic
coverage in certain
acquisition modes

Landsat

Long-term
multispectral image
archive (15

60 m)

Environmental
monitoring,
urbanization
studies

Historical
continuity,
long-term data
availability

Low image update
frequency

OpenStreetMa
p

Crowdsourced
vector dataset of
infrastructure

and

geographic features

Urban planning,
navigation,
integration
with raster data

Fast

updates,

wide coverage,
contextual data
enrichment

Possible
inconsistencies,
incomplete
coverage due to
unregulated input

Despite major advancements in GIS and remote
sensing, traditional geospatial data processing
approaches still face several key limitations:

1.

Low accuracy and inconsistency. Manual workflows
and

classical

algorithms

often

lead

to

misclassifications in land cover analysis, potentially
causing resource misallocation and hindering the
monitoring of critical phenomena such as illegal
logging or unauthorized construction [7].

2.

High labor and time intensity. Traditional analytical
methods require significant expert involvement
and time, limiting their usefulness in fast-changing
environmental contexts where real-time insights
are crucial [9].

3.

Lack of adaptability to multidisciplinary data.
Conventional models often assume statistical
stationarity of features, which reduces their
effectiveness when integrating diverse sources
such as multispectral and hyperspectral imagery,
LiDAR point clouds, cadastral records, and socio-
economic attributes. The absence of calibration or
self-learning mechanisms for changing data
distributions restricts the detection of subtle

spatiotemporal patterns and lowers predictive
performance in highly dynamic environments [6].

Thus, the current phase of GEOINT development is
marked by a shift from manual, conventional methods
toward integrated digital solutions that fuse
multimodal data sources with advanced analytics.
Overcoming the identified challenges paves the way for
improved

accuracy,

responsiveness,

and

interpretability

essential for effective resource

management and decision-making in a rapidly changing
world.

2. Machine Learning in Geospatial Analysis and
Intelligent Search

Recent advances in machine learning (ML) are having a
transformative impact on geospatial analysis,
significantly enhancing the efficiency and precision of
information extraction from large-scale datasets. In
particular, deep neural networks and anomaly
detection

algorithms

have

become

essential

components of modern GEOINT systems, expanding
the capabilities of intelligent search. The integration of
these methods enables automated image classification,
semantic segmentation, and pattern recognition, all of
which are critical for improving query relevance and


background image

The American Journal of Engineering and Technology

105

https://www.theamericanjournals.com/index.php/tajet

decision-making accuracy [4, 5].

Among the most widely used and effective tools in
geospatial analysis are convolutional neural networks
(CNNs), which have demonstrated strong performance
in

processing

satellite

imagery. The

VGG16

architecture, for instance, is commonly employed for
image classification tasks and provides high accuracy in
identifying land cover types and infrastructure
elements. In parallel, segmentation models such as U-
Net offer detailed pixel-level annotation, which is vital
for defining object boundaries and analyzing
environmental change [1].

Traditional techniques often fall short when it comes to
detecting rare events and unexpected changes in
geospatial data. In such cases, anomaly detection
algorithms like Isolation Forest and One-Class SVM are
especially useful for identifying unusual patterns. These
methods enable the detection of land cover
disruptions, unauthorized construction, and other
anomalies that may influence analytical outcomes and
the timeliness of operational decisions [6].

Modern search systems aim not only to retrieve
information but also to conduct deep analytical
processing, which necessitates the use of machine

learning techniques. The integration of natural
language processing (NLP) algorithms and knowledge
graph construction supports contextual semantic
enrichment of search results, improving both accuracy
and interpretability [2, 7]. NLP technologies, in
particular, allow systems to analyze and structure
informal text data and link it to geographic information,
creating comprehensive models for intelligent search
that align with user intent [8].

The application of deep learning in geospatial analysis
and its integration with intelligent search technologies
opens new pathways for building advanced analytical
systems. By combining high-quality satellite imagery
with powerful computational models, it becomes
possible to accelerate responses to environmental
changes, improve monitoring accuracy, and optimize
decision-making processes. This interdisciplinary
approach forms the foundation for innovative solutions

capable of addressing the complex demands of today’s

data-driven landscape.

To provide a clearer understanding of the comparative
characteristics of models used in geospatial analysis
and intelligent search, Table 2 presents a summary
comparison.

Table 2. Comparative analysis of machine learning models for geospatial analysis and intelligent

search [1, 2, 6]

Model

Task Type

Primary Application

Advantages

Limitations

VGG16

Image

classification

Land cover

identification,
infrastructure

detection

High accuracy,

strong feature

extraction

Computationally

intensive, requires

significant resources

U-Net

Semantic

segmentatio

n

Pixel-level annotation

of satellite images

Accurate

segmentation, local

and global feature

learning

Requires large

training datasets,

sensitive to tuning

Isolation

Forest

Anomaly

detection

Detection of structural

anomalies,

environmental change

Effective on sparse

anomalies, fast

computation

Can yield false

positives with

complex data

structures

One-

Anomaly

Identification of rare

Flexible

Sensitive to


background image

The American Journal of Engineering and Technology

106

https://www.theamericanjournals.com/index.php/tajet

Model

Task Type

Primary Application

Advantages

Limitations

Class

SVM

detection

events, change

monitoring

configuration,

versatile across data

types

hyperparameters,

computationally

heavy at scale

NLP

models

(e.g.,

BERT)

Semantic

information

extraction

Context-aware search

enrichment, knowledge

graph construction

Deep text

understanding,
integrable with

diverse sources

Requires large

annotated corpora for

training

In conclusion, the application of machine learning in
geospatial analysis and intelligent search not only
demonstrates high effectiveness in classification and
segmentation tasks but also opens new avenues for the
integration of multimodal data. This leads to the
development of more precise, adaptive, and
interpretable information retrieval systems. The
combined use of these technologies expands the
capabilities of analytical platforms, enabling timely
detection of environmental changes and improving the
quality of search outcomes

an essential advancement

for both applied and theoretical research.

3. Integration of Geospatial Data and Intelligent
Search: Opportunities and Prospects

Modern geospatial intelligence (GEOINT) systems are
increasingly adopting intelligent search methods to
extract meaningful insights from heterogeneous data
sources. Integrating geospatial data

including satellite

imagery, vector formats, and crowd-generated
content

with intelligent search algorithms such as

natural language processing, knowledge graphs, and
multimodal models opens new frontiers for advanced
analytics. These capabilities have the potential to
significantly enhance decision-making in domains such
as environmental monitoring, national security, and
urban planning [3].

Combining geospatial data with intelligent search
systems creates a synergy between visual and textual
information, enabling:

Contextual enrichment of search results. The
addition of spatial features enhances the depth of
query interpretation and enables geographic
context to inform ranking and retrieval [2].

Improved accuracy and relevance. The fusion of
high-resolution satellite imagery (e.g., Sentinel-2,
Landsat) with NLP-driven insights (e.g., BERT-based
models) enables more comprehensive and precise
information extraction.

Accelerated data processing. Leveraging cloud
platforms and parallel computing allows for near
real-time analysis of large-scale datasets, which is
critical for time-sensitive decision-making in rapidly
changing environments [1].

The scientific and technical potential of integrating
geospatial data and intelligent search rests on
several key pillars:

Development of multimodal models. Unifying
textual, visual, and vector data in a single analytical
framework enhances model interpretability and
analytical

performance

[2].

For

example,

architectures that combine CNNs for image analysis
with NLP modules for semantic understanding
demonstrate notable advantages over unimodal
approaches.

Knowledge graph integration. Linking geospatial
data with external knowledge sources through
semantic graphs supports the construction of
context-rich models capable of capturing deep
relationships between entities

an essential

feature for intelligent search applications [8].

Implementation of flexible and adaptive systems.
Current research focuses on designing systems that
can continuously update their models based on
incoming data. Techniques such as transfer
learning and data fusion promote model
adaptability to evolving conditions and user


background image

The American Journal of Engineering and Technology

107

https://www.theamericanjournals.com/index.php/tajet

requirements

an increasingly vital aspect of

GEOINT workflows [6, 10].

Table 3. Comparative analysis of geospatial data integration and intelligent search approaches [1, 3, 6, 10]

Approach

Description

Primary

Application

Advantages

Limitations

Data
Fusion

Integration of raster (satellite
imagery) and vector data
(e.g., OSM)

Environmental
monitoring,
urban analysis

Enhanced
detail,

richer

contextual
information

Data
harmonization
challenges,
potential
inconsistencies

Knowledge
Graph
Integration

Creation of semantic graphs
linking geospatial entities to
information sources (e.g.,
NLP, knowledge bases)

Improved
interpretability
and

search

precision

Deep semantic
connectivity,
hidden
relationship
discovery

High
computational
demands, need
for

frequent

updates

Multimoda
l Models

Integration of image, text,
and vector data in a unified
analytical model

Complex
analytics,
forecasting,
adaptive
decision-
making

Synergy across
data

types,

improved
model
accuracy

Requires large
labeled
datasets,
complex

to

develop

and

train

In conclusion, integrating geospatial datasets with
advanced semantic search mechanisms unlocks new
opportunities for higher-quality analytics through the
combination of precise spatial context and intelligent
information extraction. Building unified ecosystems
that connect diverse geodata sources with machine
learning architectures allows for the creation of
enriched spatiotemporal representations. These, in
turn, enhance pattern analysis and enable real-time
responsiveness to changing conditions.

The use of hybrid models

such as combining graph

neural networks

to capture complex entity

relationships with transformers for semantic indexing
of

textual

descriptions

delivers

high-precision

forecasts. To overcome implementation challenges,
cloud platforms with microservice-based processing
and dynamic resource allocation are recommended.
Adaptive calibration mechanisms allow real-time

tuning of algorithms to current data characteristics,
reducing preprocessing overhead.

Ultimately, these solutions expand foundational
research capabilities while laying the groundwork for
automated decision-support systems that can
effectively respond to evolving external conditions and
user needs.

CONCLUSION

The analysis conducted in this study demonstrates that
the integration of heterogeneous geospatial sources
with modern machine learning techniques significantly
enhances the functional capabilities of intelligent
search platforms. The proposed methodology is built
on a cross-modal framework that combines satellite
imagery, vector-based knowledge systems, and deep
learning architectures.


background image

The American Journal of Engineering and Technology

108

https://www.theamericanjournals.com/index.php/tajet

At the same time, several technical and methodological
challenges were identified, including the need for
robust alignment and normalization algorithms for
heterogeneous datasets, the high computational
demands of training deep models, and the limited
adaptability of current systems in rapidly changing
contexts. Future directions include the development of
transfer learning techniques and multi-level data
fusion, as well as the creation of dynamic, self-adjusting
architectures capable of responding to evolving user
requirements and environmental conditions.

In summary, the integrative approach presented here
not only addresses existing gaps in GEOINT-related
research but also opens up substantial opportunities
for the deployment of such technologies in strategic
domains

ranging from environmental monitoring and

urban planning to national security applications.

REFERENCES

Kolluru V. et al. Geospatial Intelligence Enhancement
Using Advanced Data Science and Machine Learning: A
Systematic Literature Review //Available at SSRN
4929468.

2024.

pp.1-19.

Ghadge N. Machine Learning: Enhancing Intelligent
Search and Information Discovery.

2024.

pp.1-9.

Breunig M. et al. Geospatial Data Management
Research: Progress and Future Directions //ISPRS
International Journal of Geo-Information.

2020.

Vol.

2 (9).

pp. 95.

Kolluru V., Mungara S., Chintakunta A. N. Combating
Misinformation With Machine Learning: Tools for
Trustworthy News Consumption //Machine Learning
and Applications: An International Journal.

2020.

Vol. 4 (7).

pp. 10.

Mungara S., Koganti S., Chintakunta A. N., Kolluru V. K.,
Nuthakki Y. Exploring Consumer Behaviors in E-
Commerce Using Machine Learning. International
Journal of Data Analytics Research and Development
(IJDARD). - 2023. - Vol. 1(1). - pp. 51-63.

Feldmeyer D. et al. Using OpenStreetMap Data and
Machine Learning to Generate Socio-Economic
Indicators //ISPRS International Journal of Geo-
Information.

2020.

Vol.9 (9).

pp. 498.

Gromny E. et al. Creation of Training Dataset for
Sentinel-2 Land Cover Classification //Photonics
Applications in Astronomy, Communications, Industry,
and High-Energy Physics Experiments 2019.

SPIE,

2019.

Vol. 11176.

pp. 998-1006.

Páez A. The Pragmatic Turn in Explainable Artificial
Intelligence (XAI) //Minds and Machines.

2019.

Vol.

3 (29).

pp. 441-459.

Wang J. et al. Thirty Years of Machine Learning: The
Road to Pareto-Optimal Wireless Networks //IEEE
Communications Surveys & Tutorials.

2020.

Vol. 3

(22).

pp. 1472-1514.

Bhatt S. et al. Knowledge Graph Semantic Enhancement
of Input Data for Improving AI //IEEE Internet
Computing.

2020.

Vol. 2 (24).

pp. 66-72.

References

Kolluru V. et al. Geospatial Intelligence Enhancement Using Advanced Data Science and Machine Learning: A Systematic Literature Review //Available at SSRN 4929468. – 2024. – pp.1-19.

Ghadge N. Machine Learning: Enhancing Intelligent Search and Information Discovery. – 2024. – pp.1-9.

Breunig M. et al. Geospatial Data Management Research: Progress and Future Directions //ISPRS International Journal of Geo-Information. – 2020. – Vol. 2 (9). – pp. 95.

Kolluru V., Mungara S., Chintakunta A. N. Combating Misinformation With Machine Learning: Tools for Trustworthy News Consumption //Machine Learning and Applications: An International Journal. – 2020. – Vol. 4 (7). – pp. 10.

Mungara S., Koganti S., Chintakunta A. N., Kolluru V. K., Nuthakki Y. Exploring Consumer Behaviors in E-Commerce Using Machine Learning. International Journal of Data Analytics Research and Development (IJDARD). - 2023. - Vol. 1(1). - pp. 51-63.

Feldmeyer D. et al. Using OpenStreetMap Data and Machine Learning to Generate Socio-Economic Indicators //ISPRS International Journal of Geo-Information. – 2020. – Vol.9 (9). – pp. 498.

Gromny E. et al. Creation of Training Dataset for Sentinel-2 Land Cover Classification //Photonics Applications in Astronomy, Communications, Industry, and High-Energy Physics Experiments 2019. – SPIE, 2019. – Vol. 11176. – pp. 998-1006.

Páez A. The Pragmatic Turn in Explainable Artificial Intelligence (XAI) //Minds and Machines. – 2019. – Vol. 3 (29). – pp. 441-459.

Wang J. et al. Thirty Years of Machine Learning: The Road to Pareto-Optimal Wireless Networks //IEEE Communications Surveys & Tutorials. – 2020. – Vol. 3 (22). – pp. 1472-1514.

Bhatt S. et al. Knowledge Graph Semantic Enhancement of Input Data for Improving AI //IEEE Internet Computing. – 2020. – Vol. 2 (24). – pp. 66-72.