Beginning Anomaly Detection Using

Figure 8-15.  Object-detecting video surveillance system

Chapter 8   praCtiCal Use Cases of anomaly DeteCtion

---

314

can collect all the sensor output from the tens of thousands of sensors attached to the 

tens of thousands of components, then it becomes possible for us to collect such data 

for a longer period of time and train sophisticated anomaly detection models such as 

autoencoders, LSTMs, and TCNs.

Figure 

8-16

 shows a manufacturing plant with sensor readings.

Figure 8-16.  Manufacturing plant with sensor readings

Chapter 8   praCtiCal Use Cases of anomaly DeteCtion

---

315

 Smart  Home

Another kind of business that is also using anomaly detection to its advantage is the 

smart home system. Smart homes have lots of integrated components, such as smart 

thermostats, refrigerators, and interconnected devices, that all talk to each other. 

Let’s say you have an Amazon Alexa. Alexa can talk to your smart lights, which use 

smart bulbs. All components can use a very smart app on your smart phone. Even 

thermostats are interconnected. So how do we really use anomaly detection in this 

use case? A simple way is to monitor how you set your thermostat for the optimal 

temperature during all weather conditions and follow some sort of recommendation 

or recommended behavior. Because the thermostats are personalized to some extent 

in each household, there may be a very good deep learning algorithm out there that is 

continuously looking for the thermostats across all houses, including yours, and can 

then detect how you use it normally. Figure 

8-17

 is an illustration of a smart home.

 Retail

Another big industry that uses anomaly detection algorithms is the retail industry. In 

the retail industry, there are certain use cases such as the efficiency of the supply chain 

in terms of distribution of goods and services. Also interesting are the returns from 

customers because returned goods are tricky: sometimes it costs less to sell them in a 

clearance sale than to restock.

Figure 8-17.  A smart home

Chapter 8   praCtiCal Use Cases of anomaly DeteCtion

---

316

Looking at customer sales is also critical both in terms of revenue generated by 

sales and in terms of planning future products and sales strategies, especially when it 

comes to targeting the consumers better. Figure 

8-18

 shows the historical sales figures 

of a product.