The next stage involves model training, where ML algorithms
are trained on historical data to develop predictive models. These models can
then be applied to new data for real-time predictions and decision support. The
final steps in the workflow involve feedback and optimization, where model
performance is continuously evaluated and improved based on new data and
outcomes. This integrated approach enables the development of sophisticated
predictive analytics systems that can provide actionable insights for farmers
and agricultural stakeholders. By combining the real-time data collection
capabilities of IoT with the predictive power of ML, smart farming systems can
offer unprecedented levels of precision and efficiency in agricultural
operations.
2.5
Literature Review
Numerous studies have explored the integration of IoT and
ML for predictive analytics in smart farming, contributing to a growing body of
knowledge in this field. Liakos et al. (2018) provided a comprehensive review
of machine learning applications in agriculture, highlighting the potential of
ML algorithms in various aspects of farming, including crop and soil
management, livestock production, and water management. Their work underscored
the versatility of ML techniques in addressing diverse agricultural challenges.
Goap et al. (2018) proposed an
IoT-based smart irrigation system that uses soil moisture sensors and weather
forecast data to optimize water usage. The system employed a decision tree
algorithm to predict soil moisture levels and determine irrigation schedules.
This study demonstrated the practical application of integrated IoT and ML
technologies in addressing one of the most critical aspects of agriculture -
water management. Isabelle et al. (2019) developed a machine learning-based
approach for early detection of plant diseases using hyperspectral imaging.
Their method combined IoT sensors for data collection with convolutional neural
networks for image classification, achieving high accuracy in identifying
various plant diseases. This research highlighted the potential of advanced
imaging techniques and deep learning in crop health monitoring.
Chlingaryan et al. (2018)
reviewed machine learning techniques for crop yield prediction, emphasizing the
importance of integrating multiple data sources, including IoT sensors,
satellite imagery, and historical yield data, to improve prediction accuracy.
Their work underscored the complexity of yield prediction and the need for
sophisticated ML approaches that can handle diverse data types. Wolfert et al.
(2017) examined the role of big data in smart farming, discussing the
challenges and opportunities associated with integrating IoT and ML
technologies in agricultural decision-making processes. Their research
highlighted the transformative potential of data-driven agriculture while also
acknowledging the barriers to adoption and implementation.
These studies, among others, demonstrate the growing
interest and potential of IoT and ML integration in smart farming. However,
they also highlight the need for further research to address challenges related
to data quality, scalability, and the development of robust predictive models
tailored to specific agricultural contexts. As the field continues to evolve,
ongoing research will be crucial in refining techniques, overcoming challenges,
and realizing the full potential of IoT and ML integration in agriculture.
3. Techniques in IoT and ML Integration for Smart Farming
The integration of IoT and ML in smart farming involves a
range of techniques for data collection, processing, and analysis. This section
explores the key techniques employed in each stage of the integration process,
providing a comprehensive overview of the technological ecosystem that supports
predictive analytics in agriculture.
3.1
IoT Data Collection Techniques
At the foundation of smart farming lies the collection of
data through IoT sensor networks. These networks form the backbone of data
collection, consisting of various types of sensors deployed across agricultural
fields or livestock facilities. Soil sensors play a crucial role in measuring
parameters such as moisture content, temperature, pH, and nutrient levels. This
information is vital for optimizing irrigation and fertilization practices,
ensuring that crops receive the right amount of water and nutrients at the
right time. Environmental sensors complement soil sensors by monitoring air
temperature, humidity, light intensity, and CO2 level. This data helps farmers
understand the microclimates within their fields and make informed decisions
about crop management. Plant sensors take this a step further by assessing crop
health directly, measuring factors such as leaf temperature and chlorophyll
content. These sensors can provide early warning signs of plant stress or
disease, allowing for timely interventions.
In livestock farming, animal sensors track vital
information such as location, body temperature, and activity levels. This data
is invaluable for monitoring animal health, optimizing feeding strategies, and
managing grazing patterns. The sensor networks are typically organized in a
hierarchical structure, with local clusters of sensors communicating with
gateway devices that aggregate and transmit data to central servers or cloud
platforms.
The choice of wireless communication protocol for
transmitting data from IoT sensors to central systems is a critical
consideration in smart farming implementations. LoRaWAN (Long Range Wide Area
Network) has gained popularity for its ability to cover large agricultural
areas with low power consumption. This makes it particularly suitable for
remote or expansive farming operations. Zigbee, on the other hand, is often
used for dense sensor networks within smaller areas due to its short-range,
low-power characteristics. For applications requiring higher data rates or
real-time communication, cellular networks (3G/4G/5G) provide a viable
solution, especially in areas with good cellular coverage. Wi-Fi is typically
employed for short-range, high-bandwidth applications within farm buildings or
in areas where power constraints are less of a concern.
In addition to ground-based sensors, smart farming
increasingly relies on aerial and satellite imagery for comprehensive field
monitoring. Unmanned Aerial Vehicles (UAVs) or drones equipped with various
types of cameras provide high-resolution imagery that can be used for crop
health assessment, weed detection, and field mapping. Multispectral and
hyperspectral imaging techniques capture data across multiple spectral bands,
allowing for detailed analysis of crop health and stress levels. This
technology can detect issues that are not visible to the naked eye, such as
early signs of disease or nutrient deficiencies. Thermal imaging is another
powerful tool in the smart farming arsenal, particularly for irrigation
management. By detecting temperature variations in crops and soil, farmers can
identify areas of water stress and optimize irrigation practices. LiDAR (Light
Detection and Ranging) technology, often deployed on drones or ground-based
vehicles, generates high-resolution 3D maps of agricultural landscapes. These
maps provide valuable information about terrain, crop height, and biomass
estimation. Satellite imagery complements drone-based data collection by
offering broader coverage and regular monitoring capabilities. While typically
of lower resolution than drone imagery, satellite data is invaluable for
tracking large-scale patterns and changes over time. The integration of
satellite and drone imagery with ground-based sensor data provides a
comprehensive view of agricultural operations, enabling more informed
decision-making.
3.2
Data Preprocessing and Feature Extraction
The raw data collected from IoT devices often contains
noise, missing values, and outliers, necessitating robust preprocessing
techniques to ensure data quality. Data cleaning is a critical first step in
this process, involving the identification and handling of anomalous data
points. Statistical methods or machine learning algorithms can be employed to
detect outliers, which can then be removed or adjusted based on domain
knowledge and the specific requirements of the analysis. Missing value
imputation is another crucial aspect of data preprocessing in smart farming
applications. Various techniques can be applied, ranging from simple methods
like mean imputation to more sophisticated approaches such as regression
imputation or Multiple Imputation by Chained Equations (MICE). The choice of
imputation method depends on the nature of the data and the patterns of
missingness [11].
Data normalization is often necessary to bring different
features to a common scale, particularly when working with diverse sensor types
that measure various physical quantities. Techniques such as Min-Max scaling or
Z-score normalization are commonly employed to ensure that all features
contribute appropriately to subsequent analyses and machine learning models. Feature
extraction and selection techniques play a vital role in identifying the most
relevant information from the preprocessed data. Principal Component Analysis
(PCA) is a widely used technique for dimensionality reduction, helping to preserve
important information while reducing the computational complexity of subsequent
analyses. In the context of smart farming, PCA can be particularly useful for
handling high-dimensional spectral data from multispectral or hyperspectral
sensors.
which can identify cyclical patterns in time-series data,
and wavelet analysis, which is particularly useful for detecting localized
changes in time-series signals. These techniques can help extract meaningful
information from sensor data that varies over time, such as daily temperature
fluctuations or seasonal growth patterns. In the realm of remote sensing and
spectral imaging, the calculation of vegetation indices is a common feature
extraction technique. The Normalized Difference Vegetation Index (NDVI) is one
of the most widely used indices, providing a measure of plant health and vigor
based on the reflectance of red and near-infrared light. Other indices, such as
the Enhanced Vegetation Index (EVI) or the Soil-Adjusted Vegetation Index
(SAVI), can provide additional insights into crop conditions and soil
characteristics. Feature importance ranking is another crucial step in the data
preparation process. Methods such as mutual information analysis or correlation
analysis can help identify the most informative features for a given prediction
task. In the context of crop yield prediction, for example, this process might
reveal that soil moisture levels and temperature during specific growth stages
are particularly important predictors.
3.3 Machine Learning Algorithms for Predictive Analytics
The choice of machine learning algorithm for predictive
analytics in smart farming depends on the specific application and the nature
of the data. Supervised learning algorithms are commonly employed when there is
a clear target variable to predict, such as crop yield or disease presence. Artificial
Neural Networks (ANNs) have gained significant traction in agricultural
applications due to their ability to model complex, non-linear relationships.
Multi-layer perceptrons can be effective for a wide
range of prediction tasks, while more advanced deep learning models, such as
Convolutional Neural Networks (CNNs), have shown promise in image-based tasks
like crop disease detection or weed identification.
Support Vector Machines (SVMs) have proven effective for
both classification and regression problems in agriculture. Their ability to
handle high-dimensional data makes them particularly suitable for applications
involving spectral data or multiple sensor inputs. SVMs have been successfully
applied to tasks such as crop type classification, soil moisture estimation,
and crop yield prediction. Random Forests, an ensemble learning method, have
gained popularity in agricultural applications due to their robustness and
ability to handle diverse types of features. They are less prone to overfitting
compared to single decision trees and can provide insights into feature
importance. Random Forests have been effectively used for tasks such as crop
yield prediction, land cover classification, and soil property mapping.
Gradient Boosting Machines, including algorithms like XGBoost and LightGBM, have emerged as powerful tools for
high-performance prediction tasks in agriculture. These algorithms can handle
complex interactions between features and often outperform other methods in
terms of prediction accuracy. They have been successfully applied to crop yield
forecasting, pest outbreak prediction, and irrigation scheduling.While supervised learning algorithms
dominate many predictive analytics tasks in smart farming, unsupervised
learning techniques also play important roles. K-means clustering, for example,
can be used for segmenting agricultural land into homogeneous zones based on
soil properties or crop performance. This can inform precision agriculture
practices by allowing farmers to tailor management strategies to specific zones
within their fields.
Hierarchical clustering is another unsupervised technique
that can be valuable for creating taxonomies of crop varieties or soil types.
This can help in understanding the relationships between different crop
cultivars or in classifying soil types based on multiple properties. Time
series analysis and forecasting techniques are particularly relevant in
agriculture due to the seasonal nature of crop production and the importance of
weather patterns. Autoregressive Integrated Moving Average (ARIMA) models and their
variants have been widely used for forecasting crop yields, prices, and weather
conditions. More advanced techniques like Long Short-Term Memory (LSTM)
networks, a type of recurrent neural network, have shown promise in capturing
long-term dependencies in agricultural time series data.
3.4
Model Training and Validation
The process of training and validating machine learning
models for agricultural applications requires careful consideration of the
unique characteristics of agricultural data. Cross-validation techniques are
commonly employed to assess model performance and generalizability. However,
traditional k-fold cross-validation may not be appropriate for time series data
or spatially correlated agricultural data. For time series predictions,
techniques such as walk-forward validation or time series cross-validation are
often more suitable. These methods maintain the temporal order of the data
during the validation process, providing a more realistic assessment of model
performance for future predictions.
In spatial applications, such as crop yield prediction
across different fields or regions, spatial cross-validation techniques can be
employed. These methods account for spatial autocorrelation in the data,
ensuring that the model's performance is evaluated on truly independent test
sets. Ensemble methods, which combine predictions from multiple models, have
shown promise in improving the robustness and accuracy of agricultural
predictions. Techniques such as bagging, boosting, and stacking can be used to
create ensemble models that leverage the strengths of different algorithms or
capture different aspects of the underlying agricultural processes.
3.5
Interpretation and Explainability
As machine learning models become increasingly complex, the
need for interpretability and explainability in agricultural applications has
grown. Farmers and agricultural managers often require not just predictions,
but also insights into the factors driving those predictions. Feature
importance techniques, such as permutation importance or SHAP (SHapley Additive exPlanations)
values, can provide insights into which input variables are most influential in
a model's predictions. This information can be valuable for understanding the
key drivers of crop yield or identifying the most important factors in disease
susceptibility.
Partial dependence plots and individual conditional
expectation plots are tools that can visualize the relationship between input
features and model predictions. These techniques can reveal non-linear
relationships and interactions between variables, providing valuable insights
into complex agricultural systems. For image-based models, such as those used
for disease detection or crop quality assessment, techniques like
Gradient-weighted Class Activation Mapping (Grad-CAM) can highlight the regions
of an image that are most important for the model's predictions. This can help
validate the model's decision-making process and provide interpretable results
to end-users.
4. Challenges in IoT and ML Integration for Smart Farming
While the integration of IoT and ML in smart farming offers
tremendous potential, it also presents several challenges that need to be
addressed for widespread adoption and effective implementation. These
challenges span technical, economic, and social domains, requiring
interdisciplinary approaches to overcome.
4.1
Data Quality and Reliability
One of the primary challenges in IoT-based data collection
for smart farming is ensuring data quality and reliability. Agricultural
environments can be harsh, with sensors exposed to extreme weather conditions,
dust, and physical damage. This can lead to sensor malfunctions, calibration
drift, and data inconsistencies. Developing robust, weather-resistant sensors
and implementing regular maintenance and calibration protocols are essential
for maintaining data quality. Data gaps due to sensor failures or communication
issues can significantly impact the performance of machine learning models.
Sophisticated data imputation techniques and anomaly detection algorithms need
to be developed to handle missing or erroneous data effectively. Additionally,
ensuring the temporal and spatial consistency of data collected from various
sources (e.g., ground sensors, drones, and satellites) presents a significant
challenge that requires advanced data fusion techniques.
4.2
Scalability and Infrastructure
As
smart farming systems grow to cover larger areas and incorporate more diverse
data sources, scalability becomes a critical challenge. The infrastructure
required to collect, transmit, store, and process vast amounts of agricultural
data can be substantial. In many rural areas, limited internet connectivity and
unreliable power supplies can hinder the deployment of IoT devices and the
transmission of data to central processing systems. Edge computing has emerged
as a potential solution to some of these scalability challenges. By processing
data closer to its source, edge computing can reduce bandwidth requirements and
latency, enabling more responsive and efficient smart farming systems. However,
implementing edge computing in agricultural settings presents its own set of
challenges, including the need for robust, low-power edge devices capable of
running complex ML algorithms.
4.3
Interoperability and Standardization
The lack of standardization in IoT devices and data formats
poses a significant challenge to the integration of diverse agricultural
technologies. Different manufacturers may use proprietary protocols and data
formats, making it difficult to combine data from various sources or to switch
between different systems. Efforts towards developing and adopting open
standards for agricultural IoT devices and data exchange are crucial for
fostering innovation and interoperability in smart farming ecosystems.
4.4
Data Privacy and Security
As
smart farming systems collect and process increasingly detailed data about
agricultural operations, concerns about data privacy and security come to the
forefront. Farmers may be hesitant to share sensitive data about their
operations, fearing that it could be exploited by competitors or used against
them by regulators or market players. Ensuring the security of IoT devices and
data transmission channels is crucial to protect against unauthorized access
and data breaches. Developing robust data governance frameworks that clearly
define data ownership, access rights, and usage policies is essential for
building trust in smart farming systems. Blockchain technology has been
proposed as a potential solution for creating transparent and secure data
sharing mechanisms in agriculture, but its implementation at scale remains
challenging.
4.5
Model Generalization and Adaptability
Agricultural systems are inherently complex and variable,
with significant differences across geographical regions, crop types, and
management practices. Developing machine learning models that can generalize
well across these diverse conditions is a significant challenge. Models trained
on data from one region or crop type may not perform well when applied to
different contexts. Transfer learning techniques and meta-learning approaches
offer potential solutions to this challenge, allowing models to adapt more
readily to new conditions. However, these techniques are still in their early
stages of application in agriculture and require further research and
development.
4.6 Integration with Existing Farm Management Practices
The adoption of IoT and ML technologies in farming requires
significant changes to existing farm management practices. Many farmers may
lack the technical expertise to implement and maintain these systems
effectively. There is a need for user-friendly interfaces and decision support
tools that can translate complex data and model outputs into actionable
insights for farmers. Furthermore, integrating predictive analytics into
day-to-day farm operations requires careful consideration of workflow design
and change management. Developing training programs and support systems to help
farmers and agricultural professionals effectively use smart farming
technologies is crucial for their successful adoption.
4.7
Cost and Return on Investment
The initial investment required for implementing IoT and ML
systems in agriculture can be substantial, particularly for small and
medium-sized farms. The costs associated with sensors, communication
infrastructure, data storage, and analytics platforms can be prohibitive for
many farmers. Demonstrating a clear return on investment and developing
cost-effective solutions tailored to different scales of agricultural
operations are essential for widespread adoption.
5. Future Directions and Emerging Trends
As the field of smart farming continues to evolve, several
emerging trends and future directions hold promise for addressing current
challenges and further advancing the integration of IoT and ML in agriculture.
5.1
Edge AI and Federated Learning
The development of more powerful edge computing devices and
efficient ML algorithms is enabling the deployment of sophisticated AI models
directly on IoT devices in the field. This trend towards Edge AI can help
address issues of latency, bandwidth limitations, and data privacy by
processing sensitive data locally rather than transmitting it to centralized
servers. Federated learning, a technique that allows ML models to be trained
across multiple decentralized devices without exchanging raw data, is gaining
attention in the agricultural sector. This approach could enable collaborative
learning across different farms while preserving data privacy, potentially
leading to more robust and generalizable models.
5.2
Explainable AI for Agriculture
As ML models become more complex, there is a growing
emphasis on developing explainable AI systems that can provide clear rationales
for their predictions and recommendations. In agriculture, where decisions can
have significant economic and environmental impacts, the ability to understand
and trust model outputs is crucial. Research into domain-specific explainable
AI techniques for agriculture is likely to be a key focus in the coming years.
5.3
Integration of Diverse Data Sources
The integration of an increasingly diverse array of data
sources, including satellite imagery, drone-based hyperspectral imaging, social
media data, and market information, is set to enhance the predictive power of
smart farming systems. Developing advanced data fusion techniques and
multi-modal machine learning models capable of leveraging these diverse data
types will be a significant area of research.
5.4
Autonomous Agricultural Systems
The combination of IoT, ML, and robotics is paving the way
for increasingly autonomous agricultural systems. From self-driving tractors to
autonomous crop monitoring drones and robotic harvesters, these technologies
have the potential to address labor shortages and increase operational
efficiency. Research into robust control systems, safety protocols, and
human-robot interaction in agricultural settings will be crucial for the
widespread adoption of these technologies.
5.5
Climate-Smart Agriculture
As climate change continues to impact agricultural systems
worldwide, there is growing interest in leveraging IoT and ML technologies for
climate-smart agriculture. This includes developing predictive models for
climate change impacts on crop yields, optimizing resource use to reduce
greenhouse gas emissions, and enhancing the resilience of agricultural systems
to extreme weather events. Integrating climate models with agricultural IoT
data and ML algorithms presents both challenges and opportunities for future
research.
5.6
Blockchain for Agricultural Supply Chains
The application of blockchain technology in conjunction
with IoT and ML systems holds promise for enhancing transparency and
traceability in agricultural supply chains. By creating immutable records of
agricultural products from farm to table, blockchain can help address issues of
food safety, authenticity, and fair trade. Research into scalable blockchain
solutions for agriculture and their integration with existing IoT and ML
systems is likely to be an important trend.
5.7
Precision Livestock Farming
While much of the focus in smart farming has been on crop
production, the application of IoT and ML technologies in livestock farming is
an area of growing interest. From wearable sensors for animal health monitoring
to automated feeding systems and behavioral analysis, precision livestock
farming offers significant potential for improving animal welfare and
production efficiency. Developing specialized ML algorithms for processing and
interpreting livestock-related data will be a key area of future research.
6. Conclusion
The integration of IoT and Machine Learning for predictive
analytics in smart farming represents a transformative approach to addressing
the challenges faced by the global agricultural sector. By leveraging real-time
data collection through IoT devices and advanced analytics through ML
algorithms, smart farming systems offer the potential for significant
improvements in productivity, sustainability, and resilience of agricultural
operations. This research article has provided a comprehensive overview of the
techniques, challenges, and future directions in this rapidly evolving field.
We have explored the various IoT data collection methods, data preprocessing
techniques, and machine learning algorithms that form the backbone of
predictive analytics in smart farming. The challenges discussed, including data
quality issues, scalability concerns, and the need for interpretable models,
highlight the complexity of implementing these technologies in real-world
agricultural settings [12].
Looking to the future, emerging trends such as Edge AI,
federated learning, and the integration of diverse data sources promise to
address many of the current limitations and open up
new possibilities for smart farming applications. The potential for
increasingly autonomous agricultural systems and the application of these
technologies to address climate change impacts on agriculture underscore the
transformative potential of IoT and ML integration in the agricultural sector [13]. However, realizing this
potential will require continued research and development efforts, as well as
collaboration between technologists, agricultural scientists, farmers, and
policymakers. Addressing the technical challenges while also considering the economic,
social, and environmental implications of these technologies will be crucial
for their successful and sustainable implementation.
As we move forward, it is clear that the
integration of IoT and ML in smart farming will play a vital role in shaping
the future of agriculture. By enabling more precise, efficient, and sustainable
farming practices, these technologies have the potential to contribute
significantly to global food security and environmental sustainability in the
face of growing challenges [14]. The ongoing advancement and
adoption of smart farming technologies will be essential in creating a more
resilient and productive agricultural sector capable of meeting the needs of a
growing global population in an era of climate change and resource constraints [15].
References
[1] M. Kamal and T. A. Bablu, “Mobile
Applications Empowering Smallholder Farmers: An Analysis of the Impact on
Agricultural Development,” ijsa, vol.
8, no. 6, pp. 36–52, Jun. 2023.
[2] T. Zhou and J. Zhang, “Design and
implementation of agricultural Internet of things system based on Aliyun IoT
platform and STM32,” J. Phys. Conf. Ser.,
vol. 1574, no. 1, p. 012159, Jun. 2020.
[3] B. Ouyang et
al., “Initial development of the hybrid aerial underwater robotic system
(HAUCS): Internet of things (IoT) for aquaculture farms,” IEEE Internet Things J., vol. 8, no. 18, pp. 14013–14027, Sep.
2021.
[4] I. Ahmad, Z. Wan, and A. Ahmad, “A big data
analytics for DDOS attack detection using optimized ensemble framework in
Internet of Things,” Internet of Things,
vol. 23, no. 100825, p. 100825, Oct. 2023.
[5] J. Chandiramani and S. Nayak, “Big data
analytics and internet of things for urban transportation,” in Advances in Civil and Industrial Engineering,
IGI Global, 2019, pp. 244–277.
[6] Y. Suhas, V. Kohli, S. Ghaffar, and R.
Kashef, “Network flow classification and volume prediction using novel ensemble
deep learning architectures in the era of the internet of things (IoT),” in 2021 IEEE Intl Conf on Dependable, Autonomic
and Secure Computing, Intl Conf on Pervasive Intelligence and Computing, Intl
Conf on Cloud and Big Data Computing, Intl Conf on Cyber Science and Technology
Congress (DASC/PiCom/CBDCom/CyberSciTech), AB, Canada, 2021.
[7] S. Alam, “Deep Learning Applications for
Residential Energy Demand Forecasting,” AI,
IoT and the Fourth Industrial Revolution Review, vol. 14, no. 2, pp. 27–38,
2024.
[8] L. Klerkx, E. Jakku, and P. Labarthe, “A
review of social science on digital agriculture, smart farming and agriculture
4.0: New contributions and a future research agenda,” NJAS - Wageningen Journal of Life Sciences, vol. 90–91, p. 100315,
Dec. 2019.
[9] A. Tzounis, N. Katsoulas, T. Bartzanas, and
C. Kittas, “Internet of Things in agriculture, recent advances and future
challenges,” Biosystems Eng., vol.
164, pp. 31–48, Dec. 2017.
[10] M. Stočes, J. Vaněk, and J. Masner,
“Internet of things (iot) in agriculture-selected aspects,” Agris on-line Papers in, 2016.
[11] Y. Fan, “Research on feature extraction of EEG
signals using MSE-PCA and sleep staging,” in 2018 IEEE International Conference on Signal Processing, Communications
and Computing (ICSPCC), Qingdao, 2018.
[12] A. Syskov, V. Borisov, V. Tetervak, and V.
Kublanov, “Feature extraction and selection for EEG and motion data in tasks of
the mental status assessing - pilot study using Emotiv EPOC+ headset signals,”
in Proceedings of the 11th International
Joint Conference on Biomedical Engineering Systems and Technologies,
Funchal, Madeira, Portugal, 2018.
[13] S.-M. Zhou, J. Q. Gan, and F. Sepulveda,
“Classifying mental tasks based on features of higher-order statistics from EEG
signals in brain–computer interface,” Inf.
Sci. , vol. 178, no. 6, pp. 1629–1640, Mar. 2008.
[14] J. Wan, P. Zheng, H. Si, N. N. Xiong, W. Zhang,
and A. V. Vasilakos, “An artificial intelligence driven multi-feature
extraction scheme for big data detection,” IEEE
Access, vol. 7, pp. 80122–80132, 2019.
[15] J.-H. Chen, Z.-Q. Zhao, J.-Y. Shi, and C. Zhao,
“A New Approach for Mobile Advertising Click-Through Rate Estimation Based on
Deep Belief Nets,” Comput. Intell.
Neurosci., vol. 2017, p. 7259762, Oct. 2017.
