Internet of Things

MLcan quickly identify an anomaly through its “learning” and notify the users’ family in case of

any health issue. For entertainment, they can be used for instance like the Google Glass to show

relevant content by going through the past history of their owners.

Smart Cities

IoT has been heavily used by governments for envisioning smart cities. This includes utility

management, automated transportation, surveillance, and environmental monitoring. Some

of its real-life examples are.

• IoT-based sensors can help in the identification of parking slots over the city. Similarly,

ML can learn about the patterns of parking and provide insights for opportunities,

thereby reducing the parking conundrum for metropolitans.

• To address trac congestion, IoT-based sensors can use ML to identify the busiest city

points for dierent timings. Thus, these patterns can help drivers to plan their daily

journeys and improve them considerably.

• IoT-based equipment can monitor the environment and utilize machine learning to

predict any unforeseen pollution issue.

MACHINE LEARNING MODELS

One of the jargons that you would continuously come across in the Machine Learning realm is

the “model”. As you know, Machine Learning involves learning via large datasets. The bigger the

dataset is, the better ML becomes, that is, it is more synced with the reality. You can think of the

use of ML datasets as “modeling reality”.

Humans are naturally not gifted with the ability to go over millions of samples and calculate

the possibilities and predictions from these samples. For example, if you analyze the results of

the last 10 years of global economics, can you calculate and identify patterns to predict the

next economic crash? Obviously, you cannot. This is why we are training computers to learn.

Consider a dataset as the representation of reality. Now if we divide our dataset completely, then

each record is a representation of reality.

To understand further, suppose we have to calculate the possibility of diabetes. Now our

dataset stores data of the patient history and other multiple factors. By picking an individual sample

from our project, we have a person. Now, they can have attributes like their age (attribute) is more

than 50, their gender (attribute) is female, and they have a family history (attribute) of diabetes.

Another sample can have an individual who is less than 50, is male, and has no prior

family history for diabetes. Similarly, you can have the following listed table. Now, we will

add the most important attribute in our column that is, if the person contracted diabetes. As

ML predictions are based on historical data, therefore it is necessary to get this knowledge.

Gender <50 Family History Diabetes

F Yes No No

M No Yes Yes

F Yes No Ye s

M No Yes No

M No Yes Yes

F Yes No No

Chapter 10 Data Analytics and Machine Learning for IoT

253

Internet_of_Things_CH10_pp249-270.indd 253 9/3/2019 10:15:56 AM

In this table, we have supposedly presented a complete representation of reality. Now, we can

just use the if-statement to make a condition like the following.

if (gender = = F) && age < 50 && history = = yes){

diabetes = true;

}

This means that by going over the attributes of the table, our system follows a rule where

only those patients are diabetic who are female, are aged less than 50, and have a family history

of diabetes.

Win or Lose

Per

centage of votes

Election Candidat

e

Election Candidat

e

CLASSIFICATION

REGRESSION

Load

Sensor

Data

Extract

Features

Sensor Data

Aggregation

Platform

Machine Learning Process

Develop

Model

Test

Model

Predict

Using

Model

254 Internet of Things

Internet_of_Things_CH10_pp249-270.indd 254 9/3/2019 10:15:57 AM