It is important to develop an understanding of the situations where you can potentially apply multiclass classification. The most basic requirement is a data set with multiple predefined target classes and input features that explain the target class in a singular trend. You must know this from binary logistic regression, except that there were only two target classes.

Now, hear Ankit as he walks you through some practical examples where multiclass classification can be applied.

So, there are many practical applications of multiclass classification where you can predict classes based on the current set of information. In the video, Ankit discussed three examples, one each, from banking, e-commerce, and micro-lending industries.

In the first example, a company wants to understand the future engagement level of the customers based on their activities in the first three months and the information furnished in the application. The data set contains information furnished in the application and on the initial activity of existing old customers and their current engagement level, as shown.

The current â€˜engagement levelâ€™ of these customers has three categories i.e. high, medium, and low. So, we need to classify the current customers into these three categories. Using this, you can build a model that predicts the future engagement level of relatively new customers as high, medium, or low.

Since there are more than two target variables, we cannot use a logistic regression model to classify the data points directly.

Similarly, in the second and the third examples, classification models are built, respectively, to explain transaction type in e-commerce and to predict risk category in a micro-lending company. You can take a look at both the data sets shown below.

The image below contains the data samples for the loan company example(fintech industry).

Since both the examples above have more than two target variables, we cannot use a logistic regression model to classify the data points directly.