|
Figure 8-5. Depiction of credit card fraud
Chapter 8 praCtiCal Use Cases of anomaly DeteCtion
303
Environmental
When it comes to environmental aspects, anomaly detection has several applicable
use cases. Whether it is deforestation or melting of glaciers, air quality or water quality,
anomaly detection can help in identifying abnormal activities. Figure
8-6
is a photo of
deforestation.
Figure 8-6. Deforestation
Source: commons.wikimedia.org
Let’s look at an example of the air quality index. The air quality index provides
some kind of measurement of breathable air quality, which can be measured by using
various sensors placed at various locations in the region. These sensors measure and
send periodic data to be collected by a centralized system where such data is collected
from all of the sensors. This becomes a time series, with each measurement consisting
of several attributes or features. With each point in time having a certain number of
features, which can then be input into a neural network such as an autoencoder, we can
build an anomaly detector. Of course, we can use a LSTM or even TCN to do the same.
Figure
8-7
shows the air quality index in Seoul in 2015.
Chapter 8 praCtiCal Use Cases of anomaly DeteCtion
304
Healthcare
Healthcare is one of the domains that can benefit a lot from anomaly detection, whether
it is to prevent fraud, detect cancer or chronic illness, improve ambulatory services, etc.
One of the biggest use cases for anomaly detection in healthcare is to detect
cancer from various diagnostic reports even before there are any significant symptoms
that might indicate the presence of cancer. This is extremely important given the
serious consequences of cancer for any person. Some of the techniques in anomaly
detection that we can use here involve convolutional neural networks combined with
autoencoders.
Convolutional neural networks use the concept of dimensionality reduction to
reduce the large number of features/pixels with colors into much lower dimensionality
points using the neural networks layers. So, if we combine this convolutional neural
network with autoencoders, we can also use autoencoders to look at images such as MRI
images, mammograms, or other images from diagnostic technologies in the healthcare
industry. Figure
8-8
is a set of images from a CT scan.
Figure 8-7. Air quality index
Source: commons.wikimedia.org
Chapter 8 praCtiCal Use Cases of anomaly DeteCtion
305
Let’s look at another use case of detecting abnormal health conditions of
residents of a particular neighborhood. Typically, local hospitals are used by residents
of specific neighborhoods. Using such data, the hospital can collect and store various
kinds of health metrics from all the residents in this neighborhood. Some of the
possible metrics are blood test results, lipid profiles, glycemic values, blood pressure,
ECG, etc. When combined with demographic data such as age, sex, health conditions,
etc., this information potentially allows us to build a sophisticated AI-based anomaly
detection model.
Figure
8-9
shows different health issues observed by looking at ECG results.
Figure 8-8. CT scan images
Source: commons.wikimedia.org
Chapter 8 praCtiCal Use Cases of anomaly DeteCtion
306
There are a lot of different use cases in healthcare where we can use different
anomaly detection algorithms to implement preventative measures.
Transportation
In the transportation sector, anomaly detection can be used to ensure proper
functioning of the roadways and vehicles. If we can collect different types of events from
all the sensors that are operational on the roadways such as toll booths, traffic lights,
security cameras, and GPS signals, we can build an anomaly detection engine that we
can then use to detect abnormal traffic patterns.
Anomaly detection can also be used to look at times in schedules of public
transportation and the related traffic conditions in the similar area of transportation.
Figure 8-9. ECG results
Source: commons.wikimedia.org
Chapter 8 praCtiCal Use Cases of anomaly DeteCtion
307
We can also look for abnormal activity in terms of fuel consumption, number of
passengers the public transportation is supporting, seasonal trends, etc. Figure
8-10
is an
image of a traffic jam due to peak time unexpected traffic.
Social Media
In social media platforms such as Twitter, Facebook, and Instagram, anomaly detection
can be used to detect hacked accounts spamming everyone, false advertisements, fake
reviews, etc. Social media platforms are used extensively by billions of people, so the
amount of activity on social media platforms is extremely high and is ever growing. In
order to ensure the privacy of the individuals using the social media platforms as well
as to ensure the proper experience for each and every individual using the social media
platforms, there are many techniques that can be used to enhance the capabilities of this
system. Using anomaly detection, every individual activity can be examined for normal
and abnormal behavior.
Similarly, any advertising platforms ads, any personalized friend recommendations,
any news articles that the individual might have been interested in, such as elections,
Figure 8-10. Traffic jam
Chapter 8 praCtiCal Use Cases of anomaly DeteCtion
308
can be processed for abnormal or anomalous activity. It would be a great use case
for anomaly detection if anomaly detection could detect troll activity on your tweets,
propagandized bots, fake news, and so on. Anomaly detection can also be used to
detect if your account has been taken over, because all of a sudden your account might
be posting an immense amount of tweets, pause tweets, and comments, or might be
trolling other accounts and spamming everyone else. Figure
8-11
shows an article on
fake news on Facebook.
Do'stlaringiz bilan baham: | 
