Machine Learning Assignment -I

Recent trends in various learning techniques of machine learning and classification methods for IOT
Applications.

With the advent of rapid developments, large number of heterogeneous devices is able to connect with
the help of IOT technology. Although IOT possess very complex architecture because of connectivity of
variety of devices and services in the system.

The emergence of concepts like Data Science, Big Data, Machine Learning, IoT in recent years have
added the potential of research in today's world which is flooded with data. The use of IoT devices,
sensors, etc. which are collecting data continuously is putting a lot of pressure on the existing IoT
network which is resource-constrained.
The classification of data at the device level, edge/fog level and cloud level using machine learning
techniques.

Machine learning-based control

The next stage of IoT systems evolution is the systems with machine-learning based control when IoT
potential is used to its fullest extent. In machine learning, sensor data is continuously collected and
regularly used in standard machine learning algorithms. New models are generated, and their
applicability is then tested by analysts and / or data scientists. When the models are approved, they can
be used by an IoT system.

In a smart railway, such learning can be performed with analyzing human commands. The responses of
humans to certain sensor data are accumulated in a big data warehouse, and then the models of how to
act are built accordingly (considering the actions of humans in certain situations).

Cameras can take photos of potential problems (suspicions that there are some problems) and send
them for further analysis (either manually or with computer assistance). As soon as various images are
collected in the cloud (and the problems are identified), smart systems “learn” to determine the types of
problems without human participation and send corresponding notifications.

Machine learning potential can contribute to optimizing subway trains’ schedules. Smart system
accommodates the data about the passenger flow on different days and at different times of the day.
Then, it defines the days and the time slots when additional trains should be put on the line, and, thus,
offers schedule optimizations.

Machine Learning in IoT With IoT, devices are connected with each other, communicate with each other
and collect an enormous amount of data every day. In many applications, IoT devices may also be
programmed to trigger some actions based either on some predefined conditions or on some feedback
from the Future Internet 2019, 11, 94 5 of 23 collected data. However, in order to analyze the collected
data and extract meaningful information and create smart applications, human intervention is required.
IoT devices need not only to collect data, and communicate with other devices, but also to be
autonomous. They need to be able to take context based decisions and learn from their collected data.
This need led to the creation of the term “Cognitive IoT” (CIoT) . Also, there is a necessity for Intelligent
IoT devices, able to create automated smart applications with automated resource allocation,
communication, and network operation. Deploying ML algorithms in an IoT infrastructure can introduce
significant improvements in the applications or the infrastructure itself. ML can be applied for network
optimization, congestion avoidance, and resource allocation optimization, but also for real-time or
offline data analyzing and decision making. Moreover, as the number of devices increases, the amount
of the collected data increases as well. Having to deal with “big data” is very common in IoT applications.
Big data cannot be handled properly with conventional databases. Special infrastructure is needed to
handle the great volume of structured and unstructured data and special techniques to analyze them .
There are many ML algorithms like “Ensemble”, or Artificial Neural Networks (ANN) that can help
dealing efficiently with big data.
We could group the most important IOT applications in the following categories :
• Smart Homes: This category includes traditional home devices, such as fridges, washing machine
or light bulbs, that have been developed and are able to communicate with each other or with
authorized users via internet, offering a better monitoring and management of the devices as well
as energy consumption optimization. Apart from the traditional devices, new technologies spread,
providing smart home assistants, smart door locks, etc.
• Health-care assistance: New devices have been developed in order to improve a patient’s
well-being. Plasters with wireless sensors can monitor a wound’s state and report the data to the
doctor without the need for their physical presence. Other sensors in the form of wearable devices
or small implants, can track and report a wide variety of measurements, such as heart rate, blood
oxygen level, blood sugar level, or temperature.
• Smart Transportation: Using sensors embedded to the vehicles, or mobile devices and devices
installed in the city, it is possible to offer optimized route suggestions, easy parking reservations,
economic street lighting, telematics for public means of transportation, accident prevention,
and autonomous driving.
• Environmental Conditions Monitoring: Wireless sensors distributed in the city make the perfect
infrastructure for a wide variety of environmental conditions monitoring. Barometers, humidity
sensors, or ultrasonic wind sensors can help to create advanced weather stations. Moreover,
smart sensors can monitor the air quality and water pollution levels across the city.

• Logistics and Supply Chain Management: With the use of smart RFID tags, a product can be easily
tracked from the production to the store, reducing cost and time significantly. In addition, smart
packaging can offer features such as brand protection, quality assurance, and client personalization.
• Security and Surveillance Systems: Smart cameras can obtain video input across the streets.
With real-time visual object recognition, smart security systems can identify suspects or prevent
hazardous situations.

Machine Learning Algorithm to the Internet of Things Use

Machine learning
IoT use cases Metric to Optimize
Algorithm

Traffic Prediction,
Classification Smart Traffic
Increase Data Abbreviation

Smart Traffic, Traffic Prediction,

Clustering
Smart Health Increase Data Abbreviation
 Smart Traffic,
Traffic Prediction, Increase Data Abbreviation,
Anomaly Detection Smart
Finding Anomalies in Power Dataset
Environment

Support Vector Smart Weather

Forecasting
Regression Prediction

Economics,
Real Time Prediction,
Linear Regression Market analysis,
Reducing Amount of Data
Energy usage

Classification
Real Time Prediction,
and Regression Smart Citizens
Passengers Travel Pattern
Trees

Support Vector Classify Data,

All Use Cases
Machine Real Time Prediction

K-Nearest Passengers’ Travel Pattern,

Smart Citizen
Neighbors Efficiency of the Learned Metric

Smart
Food Safety, Passengers Travel Pattern,
Naive Bayes Agriculture,
Estimate the Numbers of Nodes
Smart Citizen

Smart City,
Outlier Detection, fraud detection,
Smart Home,
Analyze Small Data set,
K-Means Smart Citizen,
Forecasting Energy Consumption,
Controlling Air
Passengers Travel Pattern, Stream Data Analyz
and Traffic

Density-Based Labeling Data, Fraud Detection,

Smart Citizen
Clustering Passengers Travel Pattern

Reducing Energy Consumption, Forecast the

Feed Forward
Smart Health States of Elements,
Neural Network
Overcome the Redundant Data and Informatio

Principal
Monitoring
Component Fault Detection
Public Places
Analysis

Canonical Monitoring
Fault Detection
Correlation Analysis Public Places

One-class Support Smart Human

Fraud Detection, Emerging Anomalies in the da
Vector Machines Activity Control
MACHINE LEARNING TRENDS:
1. Military autonomy
Machine learning has reached a point that fully autonomous systems will soon control military
ships and even bases. Via behavioral patterning, machine learning can estimate the probability
that an approaching force is friendly or belligerent. In fact, a few ground vehicles equipped in
this manner are controlled completely by machine learning to the degree that little human
oversight is actually needed. In these instances, machine learning has powered artificial
intelligence enough that an AI-powered sentry can detect, assess, and even fire upon a threat
with deadly force. The trend here is one of humans becoming more comfortable with machines
making lethal decisions. As comfort levels rise, the number and complexity of autonomous
military units are also expected to rise.
2. Security in the home
AI-driven home-security systems are not quite common, but they are on the rise. For instance,
specific components, such as smart locks can communicate with your smartphone. However,
these systems are slated for replacement by monitoring systems that can see your home via
video, detect a threat, and notify authorities. Additionally, machine learning is predicted to
transform home security and in-home personal security in that systems will be able to predict a
threat based on
interpreting behavior, such as abuse or even kidnapping.
3. Vision
Machine learning has often been restricted to mathematical calculations, statistical analysis, and
game-based performance. However, machine learning is now able to correctly identify real-
world objects. How these objects are interpreted depends on the specific use of the robot or
software, but vision-based machine learning can identify such things as people, cats, or terrain.
Consequently, sight-powered AI is on the rise, which is expected to impact home security,
driving, and healthcare.
4. Displacement
Everyone knows that jobs involving repetitious actions are being taken over by robots, smart or
dumb. However, machine learning has also made certain white-collar professions vulnerable to
displacement. For instance, x-ray interpretation is something machine learning is making
advances in, putting the job of x-ray technicians in the cross hairs of AI displacement. Similarly,
attorneys are expected to be displaced by machine learning capable of predicting the best
pathways to winning a suit. Currently, this type of AI that predicts legal strategies is overseen by
partners who also keep a staff of attorneys on the payroll. However, as partners become
comfortable with AI decisions and as AI becomes routinely successful in making legal decisions,
jobs for interns and junior-level attorneys are predicted to decline.
5. Internet of things
Currently, devices that can connect to one another are deemed smart. However, this concept of
smart is evolving as machine learning is being applied to the so-called internet of things. For
instance, different companies have developed digital sentries that listen in on people via their
cell phones, televisions, and speakers. Alexa and Siri are two such sentries, and they are
specific to Amazon and Apple, respectively. However, a more general personality is coming in
the form of voice-based request software that will be connected to the internet of things. If you
have a coffee pot, you can ask this entity to make your coffee. If you are on your way to the
office, you can ask that your car drive you to work. As machine learning develops, the currently
segmented personalities will be connected and be able to converse and share data. The result
will be a generalized sentry that governs many of today’s routine tasks.
6. Conversation
Natural language processing (NLP) is on the rise, and it has made impressive strides that allow
machines to construct textual information based on a random initial input. In fact, one NLP
can write poetry, stories, and news articles that are disturbingly convincing. Upcoming progress
is slated to become conversational, allowing companies to serve specific needs of customers
with questions about a company’s products or services.
7. Politics and fake news
Deepfakes are on the rise, and companies and governments are bracing against the potential
confusing impact such technology will have on populations. For instance, machine learning has
reached a point that it can listen to audio data from someone and then create nuanced speech
patterns that very closely match the sound and speech patterns of the actual person.
Additionally, machine learning is becoming more adept at being able to analyze hundreds of
photographs of a single person. After analyzing the images, AI can then reconstruct video-
quality depictions of the person. The result of these two technologies is something called a
deepfake. Combined, deepfake audio and video will allow AI to construct seemingly authentic
messages from celebrities, government leaders, or even regular folk. Moreover, the technology
is expected to be entirely convincing within the next 12 months.
In terms of delivery, it is expected that such fake media will be delivered to people via social
media as social websites are currently ill equipped to detect and combat deepfakes as long as
they seem to fall within the company’s terms of service. Even when such media is finally
uncovered and removed, the intended audience will likely already be influenced in one direction
or another.