There are several types of problems that can be solved using neural networks. It is important to note that the boundaries between these types of problems are not always clear and can overlap.
Classification
Artificial neural networks are capable of sorting data based on pre-set parameters. Consider a situation where a bank needs to identify clients for a loan. The input is information about applicants, including their age, credit rating, and financial capacity.
The neural network architect data package analyzes this data and divides applicants into two categories: those who meet the criteria for a loan and those who do not. This allows the decision-making process to be automated.
Regression
Problems of this type involve estimating specific numerical values rather than defining classes.
Examples of tasks of this type:
Determining a person's age from a photograph.
Estimation of the value of a car or real estate based on its characteristics.
For example, let's say you need to determine the approximate cost of a car, given information about its year of manufacture, mileage, configuration, and other options. For this task, a neural network is provided with data on thousands of cars posted on an ad site. It analyzes this data and, based on it, suggests a suitable price for a specific car.
Case: VT-metall
Find out how we reduced the cost of attracting an application by 13 times for a metalworking company in Moscow
Find out how
Forecasting
For this type of task, a neural network analyzes a dynamic set of values and uses it to predict future changes. For example, a neural network can be used to work with:
exchange rates;
prices of oil and precious metals;
the value of shares of various companies;
the volume of traffic on the site, etc.
Clustering
Clustering has much in common with classification, but there is a key difference between the two approaches. In classification, the number of classes and the criteria for determining them are known in advance, as in the case of creditworthiness. Clustering, on the other hand, is used when there is no clear idea of the final result.
Let's look at an example. Alexey runs a large online clothing store and uses e-mail newsletters as one of his marketing tools. However, the effectiveness of such newsletters leaves much to be desired, since most recipients do not even open the advertising letters.
Email Marketing
To improve the results of the mailing, the marketer suggests making it more targeted, taking into account the characteristics and habits of each recipient. Everyone interacts with email differently:
Some people open and read emails, while others delete them immediately.
Some people click on links inside emails, and some don't.
The time for checking mail also varies.
The neural network is able to analyze the actions of all recipients and identify several groups with similar behavioral patterns. Then the marketer can develop an individual mailing strategy for each group, taking into account their characteristics and preferences.
Generation