Now, you have developed a fair understanding of situations where multiclass classification is used and how the data sets look. In this video, you will hear from Ankit as he explains the differences between binary and multi classification from a model building perspective.
- As explained in the video, in binary classification, only one classifier is required for classification. But in multi classification, more than one binary classifier is built depending upon the algorithm used (you will learn about this in detail in later segments).
- The interpretation of results is straightforward in the case of binary classification. Whatever the classifier outputs is the prediction. However, in multiclass classification, since there are several classifiers/models, interpretation is made using some algorithms and sometimes it becomes complex.
As recapped in the video, the diabetes example could be explained using a single discrete graph. However, we need multiple such graphs for cases having multiple classes. Ankit introduced an example to be used for explanation in further videos. In that example, an e-commerce firm wanted to classify their customers in preference order as shown below.
The input feature used is a matrix called ‘Purchase Score’.
Purchase score: The company’s analytical team considers parameters such as the number of purchases, total credit, product diversity, time spent on each purchase, coupons collected and the number of clicks for every customer, and scales a score of 1–5.
In the next video, Ankit will apply some dummy encoding and will show discrete classification on this data set for the three levels of preference.
So, the data set has been split into three binary columns and binary classification is applied to each of them. For example, the first graph counts First Preference as 1 and other preferences (Second/Third) as 0. The second graph counts Second Preference as 1 and other preferences as 0. Parallelly, the third graph counts Third Preference as 1 and other preferences as 0.
(Note for clarity: This graphical explanation is for illustration purposes only. In the actual model, only sigmoid curves will be classifying the data.)
By now, you must have developed a clear understanding of what and why of multiclass classification. Now, we will go ahead and solve this situation. In the next video, Ankit will introduce the techniques that you will study in this session for solving multi classification problems.
So, as Ankit explained, you will be focusing upon two multi classification techniques, i.e.,
- One vs One, and
- One vs Rest.
These are classification techniques in general, which involve various machine learning models in their implementation. In the scope of this session, you will be using logistic regression only. However, any other model can also be used on these classification techniques.
Coming Up
In the next two segments, Ankit will give a detailed diagrammatic explanation of One vs One and One vs Rest classifying techniques.
Report an error