ARTICLE IN PRESS
Machine learning and deep
learning algorithms on the
Industrial Internet of Things (IIoT)
P. Ambika
Kristu Jayanti College, Bangalore, India
Contents
1. Introduction 2
2. IIoT analytics overview 3
3. Types of data 5
3.1 Unstructured data 5
3.2 Semi-structured data 6
4. Challenges in IIoT 7
4.1 Security 7
4.2 Interoperability 7
4.3 Real-time response 7
4.4 Future readiness 8
5. Need of contextual analysis in IIoT 8
6. Machine learning for contextual analysis 9
7. Role of analytics in IIoT 10
7.1 Machine learning algorithms 10
7.2 K-nearest neighbor algorithm 11
8. Naïve Bayes algorithm 11
9. Support vector machine (SVM) 11
10. Linear regression 12
11. Random forest (RF) 12
12. K-means clustering 12
13. Principal component analysis (PCA) 13
14. Canonical correlation analysis (CCA) 13
15. Neural networks 13
15.1 Dynamic rules using machine learning and deep learning 14
15.2 Domain intelligence using ML and DL 15
16. Monitoring 16
17. Behavioral control 16
18. Optimization 16
19. Self-healing 16
20. Summary 16
Advances in Computers # 2019 Elsevier Inc. 1
ISSN 0065-2458 All rights reserved.
https://doi.org/10.1016/bs.adcom.2019.10.007
ARTICLE IN PRESS
2 P. Ambika
References 17
Further reading 18
About the author 18
Abstract
Deep transformation and human progress is a new industrial revolution that makes
“Automation of Everything.” It connects all digital interfaces, data analysis and control
of the physical world through networks of computers. This key revolution promises
everyone to unlock trillions of opportunities in the next decade. Human could feel mas-
sive improvements in productivity in physical and digital industries that enhances qual-
ity life of a human healthier and more sustainable community. In the world of IIoT, the
creation of massive amounts of data from a various sensors is common and there is lot
of challenges. This goal of this chapter is to provide a comprehensive review about
Machine learning and deep learning techniques, popular algorithms, and their impact
on Industrial Internet of Things. This chapter also delves use cases where machine learn-
ing is used and to gain insights from IoT data.
1. Introduction
Industry 4.0 revolution forcing every industry and its applications
require change that rely heavily on operational technology (OT), such as
manufacturing, transportation, energy, and healthcare. Previously, Industrial
IoT needs fog and edge computing [1] technologies to provide the necessary
connectivity within Industry 4.0. But this revolution brings another inter-
connected component that is essential for IIoT: Analytics.
Machine learning and deep learning algorithms enhance the ability of big
data analytics and IoT platforms add value to each of these segments.
Industrial IoT deals with three different types of IoT data.
1. Raw (untouched and unstructured) data.
2. Meta (data about data).
3. Transformed (valued-added data).
These algorithms will support that manage each of these data types in terms
of identifying, categorizing, and decision-making. Deep Neural Networks
coupled with big data analytics that draw meaningful and useful as informa-
tion for decision-making purposes. The use of Artificial Neural Network in
IoT and data analytics will be crucial for efficient and effective decision-
making, especially in the area of streaming data and real-time analytics asso-
ciated with edge computing networks.
Industrial Internet of Things (IIoT) applications include many industry
verticals like healthcare, retail, automotive, and transport. IIoT will signif-
icantly improve reliability, production, and customer satisfaction in many
ARTICLE IN PRESS
Machine learning and deep learning algorithms on the IIoT 3
industries. IIoT will initially improve existing processes and augmented cur-
rent infrastructure but the ultimate goal is realize entirely new and intensely
improved products and services.
Many industries understand how and where IoT technologies and solu-
tions will drive opportunities for operational improvements, new and
enhanced products and services, as well as completely new business models.
Machine learning and deep learning algorithms on IIoT will significantly
improve reliability, production, and customer satisfaction that will rely upon
integrating key technologies, devices, software, and applications. Everything
needs a substantial breadth and depth of technologies that require careful
integration and orchestration.
These technology makes machine smarter that represent intelligent
devices, machinery, equipment, and embedded automation systems [2,3]
that perform monotonous tasks and can also solve complex problem without
human intervention. It should also ensure smartness in enterprise include
improvements in the smart workplace, smart data discovery, cognitive auto-
mation and many more. A digital twin is intended to be a replica of any phys-
ical assets or processes, etc. Mostly it is referred as an outcome of internet of
things (IoT) that exponentially expanding all these devices and provide us
with an equally expanding data and can also be analyzed for its efficiency,
design, maintenance and many other factors. An important and significant
factor about any digital twin is to continuously update and “learns” in near
real-time any change that may occur. We are witnessing IoT paradigm and
its solutions made lot of changes in the industry market place.
This chapter evaluates various machine learning and deep learning
algorithms and their use relative to analytics solutions that rapidly grow in
enterprise and on industrial data arena. It also assesses emerging business
models, its solutions and provides forecasting for unit growth and revenue
for both analytics and IIoT. In particular this chapter focuses on the chal-
lenges associated with the energy efficiency, real-time performance, inter-
operability, and security. It also provides a systematic overview of the
state-of-the-art research efforts and potential research directions to solve
Industrial IoT challenges.
2. IIoT analytics overview
Industrial IoT architecture involves four key components
Things—Actual machines or systems that are involved in the industry to
be monitored by collecting their data. It is the real source of data.
ARTICLE IN PRESS
4 P. Ambika
Intelligent Edge Gateway—It is a software component that closely associ-
ated with the edge devices that is capable of collecting, aggregating, san-
itizing streaming of light data. It allows us to push aggregated and
relevant result to the IoT cloud. Generally, it acts as a mediator between
the things and the cloud IoT platform.
IoT Cloud—Core IoT platform that capable of handling enormous
amounts of data with data analytics, machine learning, and artificial intel-
ligence techniques. Device management, stream analytics, events
processing, rules engine, alerts and notifications are the processing capa-
bilities. It includes big data, machine learning capabilities and also other
services such as authentication, multitenancy, end-to-end security,
SDKs, and platform APIs.
Business Integration and Applications—It is a backend system that integrates
several IT systems to ensure machine data to complete the circle of oper-
ations. Examples of such systems are ERP, QMS, Planning and
Scheduling, etc. (Fig. 1).
Data analysis can be divided in to three types depends on the type of result it
produced. They are descriptive analysis, predictive analysis and prescriptive
analysis (Fig. 2).
Fig. 1 Ideal IIoT analytics architecture.
Fig. 2 Types of data analysis.
ARTICLE IN PRESS
Machine learning and deep learning algorithms on the IIoT 5
Descriptive data analysis defines what is happened/happening; we can
pull data to gain insight in to current operating condition. It involves sum-
marizing, organizing the data, describe the data but do not attempt to make
inferences from the sample.
Predictive analysis foresees problems before they occur. It is a branch of
advanced analytics that is used to make predictions about unknown
future events.it uses techniques from data mining, statistics, modeling,
machine learning, and AI to analyze current data to make predictions about
future.
Prescriptive analysis is one step ahead of predictive analysis and recom-
mends solutions for upcoming problems. It suggests decision options or mit-
igates a future risk, and also illustrates the implications of each decision
option. It automatically process new data to improve the accuracy of pre-
dictions and provide better decision options.
3. Types of data
Machine data does not reveal complete insight in every case. By
combining data from different sources we can create new insights. It is
important to define a few key concepts related to data and there are two cat-
egorization of data in IoT perspective whether the data is structured and
unstructured.
Structured data has a schema or model that defines how the data is orga-
nized in a simple tabular form. This type of data found in computing systems
(Fig. 3).
Structured data resides in relational database and data may be human or
machine generated as the data is created within the RDBMS structure.
Human generated queries used to extract useful information. Some of the
popular applications using structured data like railway reservation systems,
sales transactions and ATM activity.
3.1 Unstructured data
Unstructured data does not follow logical schema that decode the data
through traditional programming. This data does not fit in to a predefined
data model. The data may be textual or non-textual and human generated.
Unstructured data includes text files, email, social media, website, mobile
data, media, satellite images, scientific data, sensor data, etc.
ARTICLE IN PRESS
6 P. Ambika
Fig. 3 Structured, unstructured and semi-structured data.
3.2 Semi-structured data
Semi-structured data is a combination of structured and unstructured data
and shares characteristics of both. It also follows certain schema, consistency
and exist to ease space, clarity. CSV, XML and JSON documents are
ARTICLE IN PRESS
Machine learning and deep learning algorithms on the IIoT 7
semi-structured documents. NoSQL databases are considered as popular to
handle semi-structured data.
IIoT devices generate structured, unstructured and semi-structured data.
Structured data can be easily managed and processed because of sell defined
structure. On the other hand, semi-structured and unstructured data require
data analytics tools for preprocessing and managing. Most of the IIoT net-
works data is in transit or motion, for example, web browsing file transfers
and email. In industrial IoT perspective data from edge device passes in the
network always in motion. It may be filtered and processed by some other
devices connected in the same network or it may sent to data center. When
it comes to data center it could be processed by the real-time data analysis
software and the responses sent back to devices.
4. Challenges in IIoT
4.1 Security
Security is a challenge in IoT because of two characteristics: one is the
devices in IoT are constrained and the second is the scale. A device is con-
strained when it has limited processing power, memory, battery and band-
width. The constrained devices cannot protect themselves given their
limited computing and battery capabilities. The devices are also constrained
in terms of standard interfaces they expose for managing them. The con-
strained devices are easy target for DDoS attacks and need an external
support.
4.2 Interoperability
To address the device diversity on IoT, connected devices should follow
same language of protocols and encodings. This could be addressed by an
interoperable IoT environment. Practically it is complex in systems that
has different layers with communication protocols stack between devices.
Interoperability may also increase the economic value of the IoT market.
Purchase of IoT products and services also depends on interoperability.
Interoperability will be a significant consideration that enable connected
devices to work in a better environment.
4.3 Real-time response
Some applications require real-time response, e.g., ability control tempera-
ture of boiler. And, IoT infrastructure is still evolving, and is even more chal-
lenging in developing countries to be able to maintain connectivity to the
ARTICLE IN PRESS
8 P. Ambika
cloud all the time. An IoT application should continue to function even if
the connectivity to the cloud breaks.
4.4 Future readiness
IoT is still evolving and requires the current deployments should support
future application and new application service providers. The success of
IoT deployments depends on the ability to allow innovation over the exis-
ting infrastructure.
To overcome the above challenges and to enhance edge computing
capability, context awareness is the key component that needs to be enabled
with the use of machine learning algorithms.
5. Need of contextual analysis in IIoT
Authors said [4–6] IIoT is the result of convergence between
Operational Technology (OT) and traditional information and communi-
cation technologies (ICT). Traditional approach lacks in understanding of
the differences between conventional enterprise ICT systems and the
cyber-physical systems that are found in industrial applications. In contrast
some assessment [7] of the security challenges faced in securing IoT devices
recognized some of the constraints. It indicates current security mechanisms,
based on “traditional public-key infrastructures will almost certainly not
scale to accommodate the IoT’s amalgam of contexts and devices.”
The researchers were taking into account the limitations of IIoT devices,
for example, sensors, actuators and RFID tags, where there are processing,
power and economic constraints limiting the use of strong encryption.
Adding to that, there is also the mismatch between the relative short lifespan
of many devices and the relative longevity other devices, which are often
expected to have a lifespan an order of magnitude longer than their
counterparts.
The previous section focused on issues with IoT devices, such issues pre-
vents any device from communicating, or simply that the device’s battery
has drained and is in need of replacement/replenishment. In real-time
responsive analysis of any industrial device expected without any issues,
but there’s certain information that’s important to highlight for everyone.
For example, in agriculture devices attached to mobile agricultural
equipment (e.g., a tractor) that enable that equipment to be tracked. The
devices will likely be using GPS to get location data for the tractor, but unfor-
tunately, GPS doesn’t work effectively when vehicles are inside buildings.
ARTICLE IN PRESS
Machine learning and deep learning algorithms on the IIoT 9
If all you show on the UI is the last known location of the tractor, and
that tractor is now being stored inside a building, the GPS location won’t
reflect the actual location of the tractor and can confuse users and geo-
fences too.
In this scenario, there are no issues with the device itself. However, the
contextual state that the device is in (i.e., inside a building) is important to
the functionality of the IoT solution. Contextual data is the key component
to classify the device—and by extension, the tractor—into a state (e.g.,
“indoors”). It always provides helpful information to everyone so they’re
not confused when the device can’t get an accurate GPS position inside a
building. If the device itself can’t tell when it’s inside. All the device knows
is that it isn’t getting a good signal from GPS satellites and therefore can’t
acquire an accurate GPS position. This section explore some of the key
aspects of contextual IoT device management using machine learning and
to manage devices contextually if you’re building, buying, and/or
implementing massive-scale IoT solutions.
6. Machine learning for contextual analysis
Machine learning plays a major role in the contextual analysis in IIoT
that can help the system getting better in analyzing and troubleshooting the
network issues and issues with IoT devices, over time.
Some of the capabilities of machine learning which can be developed are:
• Prediction models
• Anomaly detection
• Recommendation models
Root Cause Analysis (RCA) capabilities aims to find the root cause of any
issues by a great accuracy by analyzing the logs and historical events. It is very
important to find out the root cause in an IoT environment so that no critical
system is out of service for long time. Manual identification of the root cause
can take hours or sometimes even days. Machine learning based Root Cause
identification system can do it in minutes or sometimes in seconds. The sys-
tem works on live data as well as on the historical data and learns over time to
get better in recognizing patterns and correlating them back to the root
cause. This is a very powerful capability which can go a long way in man-
ageability and usability of IoT deployments.
“Rule-based Analysis” can be the starting point of getting into the
machine learning territory. The most important thing is to develop a Rules
Engine, which can be decoupled from the implementation deep down.
ARTICLE IN PRESS
10 P. Ambika
This decoupling can help even the most novice users of the IoT systems to
enhance the system by adding new rules.
Security models based on these concepts can help find out and flag hid-
den and future threats. Once the system has historical data and trends, the
context feeds to itself resulting in self-learning. It is very important to train
such system for contextual analysis over a large number of network and
deployment scenarios. This would help the system giving much accurate
predictions and recommendations.
7. Role of analytics in IIoT
An Analytics System is very important for working with a very large
amount of data (big data) and creates meaningful reports and heat maps to
take decisions. The following list covers high level user visible requirements
for an analytics system:
1. Data correlation to find the overall picture
2. NBI feed in JSON format for adopters to consume the data—resulting in
better adoption and reporting. SCAP (Security Content Automation
Protocol) based reporting can also be used
3. Data funnels, filters and segmentations—important to slice and dice the
data we have
4. Both structured and unstructured data analytics—this is very important
as insights into security violation can come in any form.
All of these along with machine learning capabilities help in the contextual
analysis of the system resulting in better security.
7.1 Machine learning algorithms
Machine learning algorithms can be applied on IIoT to reap the rewards of
cost savings, improved time, and performance. In the recent era we all have
experienced the benefits of machine learning techniques from streaming
movie services that recommend titles to watch based on viewing habits to
monitor fraudulent activity based on spending pattern of the customers. It
can handle large and complex data to draw interesting patterns or trends
in them such as anomalies. Machines are needed to process information fast
and make decisions when it reaches the threshold. There are many machine
learning algorithms listed in Table 1 that help to do better data analysis in
industrial IOT devices.
ARTICLE IN PRESS
Machine learning and deep learning algorithms on the IIoT 11
Table 1 Machine learning algorithms.
Algorithm Type of the task
K-nearest neighbor Classification
Naı̈ve Bayes Classification
Support vector machine Classification
Linear regression Classification/Regression
Random forest Classification/Regression
K-means Clustering
Principal component analysis Feature extraction and dimensionality reduction
Canonical correlation analysis Feature extraction
Neural networks Classification/Regression
7.2 K-nearest neighbor algorithm
K-nearest neighbors (KNN) [8] is to classify a new given unseen data point
by looking at the K given data points in the training set that are closest to it in
the input or feature space. We use distance metric, such as Euclidean dis-
tance to find the K nearest neighbors of the new data point. To formulate
the problem, let us denote the new input vector (data point) by x, its K
nearest neighbors by Nk(x), the predicted class label for x by y, and the class
variable by a discrete random variable t.
8. Naïve Bayes algorithm
Given a new, unseen data point (input vector) z ¼ (z1, z2, …., zm),
naive Bayes classifiers, which are a family of probabilistic classifiers, classify
z based on applying Bayes’ theorem with the “naive” assumption of inde-
pendence between the features (attributes) of z given the class variable t.
Naive Bayes classifiers require a small number of data points to be
trained, and can deal with high-dimensional data points, and also are fast
and highly scalable [9]. Popular applications that use this algorithm are spam
filtering [10], text categorization, and automatic medical diagnosis [11].
9. Support vector machine (SVM)
It is probabilistic based technique that does binary classification and
aim to find the dividing hyper plane that separates both classes of the training
ARTICLE IN PRESS
12 P. Ambika
set with the maximum margin. The predicted label of a new, unseen data
point is determined based on applying machine learning algorithms to
Internet of Things use cases. Some of the application that use support vector
machine (SVM) are Anomaly Detection, Smart Traffic, Smart Environment
Traffic Prediction, Finding Anomalies in Power Dataset [12–14].
10. Linear regression
Multiple Linear Regression is a very simple approach for predicting a
quantitative response Y on the basis of multiple predictor variables Xp. In
general, suppose that we have p distinct predictors. Then the multiple linear
regression models take the form
y ¼ β0 + β1 x1 + β2 x2 + ε
11. Random forest (RF)
Random forest [15] is an ensemble learning method used for classifi-
cation similar like that build by constructing multiple of decision trees at
training time and outputting the class that is the mode of the classes (classi-
fication) or mean prediction (regression) of the individual trees.
The Random Forest model is also efficient on larger data sets, and can
handle thousands of input variables.
12. K-means clustering
K-means algorithm [16] is to cluster the unlabeled data set into K clus-
ters (groups), where data points belonging to the same cluster must have
some similarities. In the classical K-means algorithm, the distance between
data points is the measure of similarity. Therefore, K-means seeks to find a
set of K cluster centers, denoted as {s1, …., sk}, such that the distances
between data points and their nearest center are minimized. It has many
limitations because of the use of Euclidean distance as the measure of sim-
ilarity and also types of data variables that can be considered. Moreover
cluster centers are not robust against outliers. Authors [17] have used
K-means in real-time event processing and clustering algorithm for analyz-
ing sensor data that uses open IoT middleware as an interface for innovative
analytical IoT service.
ARTICLE IN PRESS
Machine learning and deep learning algorithms on the IIoT 13
13. Principal component analysis (PCA)
Principle component analysis (PCA) technique is used for feature
reduction. Actually, what PCA does is feature extraction in health analytics,
if we know that certain features are very strong in predicting Coronary
Artery Disease patients, the doctors need to know what these features are.
In order to reduce the number of feature to make the feature recommenda-
tion task tractable PCA [18] is used and it does not alter the original features.
Data is grouped in different clusters, which are usually chosen to be far
enough apart from each other spatially, in Euclidean distance, to be able to
produce effective data mining results. Each cluster has a center, called the
centroid, and a data point is clustered into a certain cluster based on how
close the features are to the centroid-means algorithm iteratively minimizes
the distances between every data point and its centroid in order to find the
most optimal solution for all the data points.
14. Canonical correlation analysis (CCA)
It is a linear dimensionality reduction technique related to PCA that
deals with one variable but CCA deals with two or more variables. Its objec-
tive is to find a corresponding pair of highly cross-correlated linear subspaces
so that within one of the subspaces there is a correlation between each
component and a single component from the other subspace. The optimal
solution can be obtained by solving a generalized eigenvector problem [19].
Authors have [20] used PCA with CCA for detecting intermittent faults and
masking failures of indoor environments.
15. Neural networks
Multi-dimensional requires a lot of computing power and it is difficult
to determine the set of parameters to input. Neural networks mimic the way
the human brain works, just like how human recognize scenes from the envi-
ronment that recollects the distinct shapes. There are different types of neural
networks and is used in many architectures, applications and use cases.
Deep Neural network types are:
• Convolutional Neural Network
• Recurrent Neural Network
• Deep Belief Network
ARTICLE IN PRESS
14 P. Ambika
Deep learning is extended effective approach on machine learning, and
exhibits its strength when a system “automatically” extracts features such
as specified colors and shapes from a massive amount of data consisting of
voice, images and other elements. Features have been extracted manually
based on a trial-and-error with the help of the knowledge provided by
experts in respective fields in most of the machine learning tasks and deep
learning allows extraction, classification and inference entirely automatically
through the learning features themselves with high accuracy. AI technology
has grown phenomenally by deep learning.
Many industries have undertaken the development of its AI technology
for many years and applying this technology into services for industrial IoT.
For example, communication AI system supports human activities by rec-
ognizing human’s intentions and situations through complex information
such as voice and images. Expanding the use of deep learning for the
industrial IoT region to improve the “speed” and “accuracy” in data analysis
utilizing experiences and knowledge cultivated with many industrial appli-
cations. In addition to voice and image data, IoT data corrected by many
kinds of sensors which are installed in industrial equipment and products
are also analyzed through machine and deep learning system which is based
on learning, inference and action. They decide to improve the quality of
customer products and services, as well as the business operations of its
customers.
Deep learning is also greatly contributing to digital twin creation. For
example, monitoring and inspecting system for electric power infrastructure
using a drone system collaborated with Alpine Electronics, Inc. and Toshiba.
Alpine Electronics, Inc. works based on image recognition technology. The
image recognition technology that uses deep learning is adopted to find the
damage on the power transmission lines. The new system allows detection of
places with a damaged part quickly with high accuracy by examining images
photographed by the drone.
15.1 Dynamic rules using machine learning and deep learning
The context is the real world scenario that arises based on the present envi-
ronment of the user. A rule is defined as a set of conditions for that context.
The real-world scenario is translated into a context that is defined by the
rules that mentions the appropriate action that is to be taken. Based on
the rules, the user takes the required decisions and triggers the action
real-time. The decision can, however, be automatic too. Once a model
ARTICLE IN PRESS
Machine learning and deep learning algorithms on the IIoT 15
Feature Engineering Anomaly detection DL Model Extraction
Collection and Streaming
aggregation IO Intensive
Framework Machine Learning Framework
framework
Human domain CPU attached storage
knowledge GPU attached storage
Deep Learning Framework
Compute
Proximity Access Intensive
Network Service Network
Network
Fig. 4 IIoT framework with ML and DL.
exists (or is created over a period of time), and the context complies with the
model, a rule-based decision is quite simple to take.
One of the ways this can be done is, to create a model over time and then
define a “baseline.” A baseline, from then onwards, can be used to compare
every data points (Fig. 4).
Few of the machine learning related requirements for such a system,
where it can be very relevant, would be as follows:
1. The system should be able to do Data Correlation using various rules.
2. The system should be able to work with different depths of analytics
based on the requirement.
3. The system should support Anomaly Detection which would come
handy when the data point does not comply with the model. In conven-
tional model, the first reaction of the system would be to discard such
data. However, a machine learning system should at least try to find
the cause of such anomaly. This can give a lot of insights into the context
where these capabilities can take corrective actions.
4. Such system should also possess the capability to perform Predictive
Analytics. This capability would keep the system ready to tackle the
upcoming situations, which can mean a threat or a condition which
requires attention.
5. The system should be able to show the Trend over a period of time.
15.2 Domain intelligence using ML and DL
Operations of ML and DL classified in to two groups, one is at the edge—
data are collected and processed locally either on the device or on the
ARTICLE IN PRESS
16 P. Ambika
gateway. Second is at a central computing unit such as cloud. Depends on
where the IIoT data is processed we can derive intelligence in to four domains.
16. Monitoring
Data on a smart object is processed based on conditions and it refers to
external factors such as temperature, humidity or presence of dioxide, etc.,
ML and DL techniques used to monitor that allows us to early detect failure
conditions.
17. Behavioral control
It always work with monitoring. A threshold value is defined along
with the given set of parameters. Machines dynamically learns and alerts
if there are deviations.
18. Optimization
Deep neural networks can be combined with the above such multiple
systems to estimate some chemical mixture for target air temperature.
Objective of such layered architecture is to optimize the result.
19. Self-healing
Popular approach in deep learning is closed loop. Systems that are
capable of doing operation optimization can also be programmable to
dynamically monitor and combine new parameters that deduce and imple-
ment new optimizations. Then the system become self-learning or healing.
20. Summary
Data is growing exponentially with the arrival of the Industrial
Internet of Things and to acquire more data, more advanced technologies
are required to scrutinize and filter out the important information and value
held within. Machine learning algorithms and deep learning techniques
allows us to extract large amounts of data, identify patterns and trends within
it, and make predictions. Industrial IoT decision-making would be a greater
advantage to optimally control and manage their assets example digital twins.
Preventing such assets from failing, forecasting demand, and optimizing
ARTICLE IN PRESS
Machine learning and deep learning algorithms on the IIoT 17
movements of the workforce are prevalent that are well managed by
machine learning techniques and Deep Neural Networks and also improve
the overall performance of the organization. By combining these elements
and more on Digital Twins would result in significant benefits.
References
[1] Aazam M, Huh E-N: Fog computing micro datacenter based dynamic resource estima-
tion and pricing model for IoT. In IEEE 29th International Conference on Advanced
Information Networking and Applications, 2015, IEEE, pp 687–694.
[2] Atzori L, Iera A, Morabito G: The internet of things: a survey, Comput Netw
54(15):2787–2805, 2010.
[3] Cecchinel C, Jimenez M, Mosser S, Riveill M: Architecture to support the collection of
big data in the internet of things. In IEEE World Congress on Services, 2014, IEEE,
pp 442–449.
[4] Steenstrup K, et al: Predicts 2017: IT and OT Convergence Will Create New Challenges and
Opportunities, 2016, Gartner.
[5] Hertzog C: Smart Grid Trends to Watch: ICT Innovations and New Entrants, 2012, OECD
Publishing. https://doi.org/10.1787/5k9h2q8v9bln-en.
[6] Shah N: IT and OT Convergence—The Inevitable Evolution of Industry, 2017, IoT for All.
[7] Roman R, Najera P, Lopez J: Securing the internet of things, IEEE Comput 44:51–58,
2011. https://doi.org/10.1109/MC.2011.291.
[8] Cover T, Hart P: Nearest neighbor pattern classification, IEEE Trans Inf Theory
13(1):21–27, 1967.
[9] McCallum A, Nigam K, et al: A comparison of event models for naive bayes text clas-
sification. In AAAI-98 Workshop on Learning for Text Categorization, Vol. 752, 1998,
Citeseer, pp 41–48.
[10] Metsis V, Androutsopoulos I, Paliouras G: Spam Filtering with Naive Bayeswhich Naive
Bayes? 2006, CEAS, pp 27–28.
[11] Webb GI, Boughton JR, Wang Z: Not so naive bayes: aggregating onedependence esti-
mators, Mach Learn 58(1):5–24, 2005.
[12] Ni P, Zhang C, Ji Y: A hybrid method for short-term sensor data forecasting in internet
of things. In 11th International Conference on Fuzzy Systems and Knowledge Discovery
(FSKD), 2014.
[13] Shukla M, Kosta Y, Chauhan P: Analysis and evaluation of outlier detection algorithms
in data streams. In International Conference on Computer, Communication and Control (IC4),
2015, IEEE, pp 1–8.
[14] Shilton A, Rajasegarar S, Leckie C, Palaniswami M: Dp1svm: a dynamic planar one-
class support vector machine for internet of things environment. In International
Conference on Recent Advances in Internet of Things (RIoT), 2015, IEEE, pp 1–6.
[15] James G, Witten D, Hastie T, Tibshirani R: An Introduction to Statistical Learning, 2013,
Springer.
[16] Coates A, Ng AY: Learning feature representations with K-Means. In Montavon G,
Orr GB, M€ uller KR, editors: Neural Networks: Tricks of the Trade. Lecture Notes in
Computer Science, Vol. 7700, Berlin, Heidelberg, 2012, Springer.
[17] Hromic H, Le Phuoc D, Serrano M, et al.: Real time analysis of sensor data for the inter-
net of things by means of clustering and event processing. In IEEE International
Conference on Communications (ICC), 2015, IEEE, pp 685–691.
[18] Naik A: k-Means Clustering Algorithm, 2016, Retrieved June 14, 2016, from https://
sites.google.com/site/dataclusteringalgorithms/k-means-clustering-algorithm.
ARTICLE IN PRESS
18 P. Ambika
[19] Bishop CM: Pattern Recognition and Machine Learning, 2006, Springer.
[20] Monekosso DN, Remagnino P: Data reconciliation in a smart home sensor network,
Expert Syst Appl 40(8):3248–3255, 2013.
Further reading
[21] Hubel D, Wiesel T: Receptive fields and functional architecture of monkey striate cor-
tex, J Physiol (Lond) 195:215–243, 1968.
About the author
Dr. P. Ambika is presently working as an
Assistant Professor in Department of
Computer Science (PG), at Kristu Jayanti
College, Bangalore. She is a Senior IEEE
Member and Secretary of IEEE Roof
Computing P1931.1 WG and she has over
12 years of experience including academics,
research and industry. She has done her
PhD in Computer Applications from Anna
University in the area of Image Processing
and Retrieval in 2014. Dr. Ambika has pub-
lished more than 12 papers in International/
National journals and conferences. She also
attended various workshops and FDPs. Her research area includes
data science and analytics, AI, machine learning, IoT and roof computing.
