"Binary classification algorithms don't work on multiclass problems" That's not true! We can use a trick to split the dataset and train binary models. Here is how it works 👇 As the name suggests binary classification works with two classes. Logistic regression and perceptron are the typical examples. But what if we don't know any other methods and we face a multi-class classification problem? We can apply a little trick. By splitting the data, we will get multiple binary classification problems. One vs All Consider this: You need to sort fruits on the table. You have Apples, Bananas, and Oranges. Instead of sorting all the fruits at once, you're going to do it one fruit at a time. You start with Apples. You create two categories: Apples and Not Apples (Oranges Bananas). After Apples, you go on to Bananas and do the same process. You will have two piles of fruits: Bananas and a mix of Apples and Oranges. You repeat the process for all categories. Do you notice what we did? We transformed a multiclass classification problem into a binary task. That's exactly how One vs All method works. For 3 categories you train 3 models: Model 1: Class 1 vs [Class 2 Class 3] Model 2: Class 2 vs [Class 1 Class 3] Model 3: Class 3 vs [Class 1 Class 2] This way the algorithm learns to recognize one category against all the others, one at a time. What happens with new data? You trained the 'model on your fruits' and you get an apple. What happens? Your 3 models will each give an output: Apple Model: 'I'm 92% sure it's an Apple.' Banana Model: 'I'm 5% sure this is a Banana.' Orange Model: 'I'm 9% sure this is an Orange.' From this output, you will choose the most confident result. And that's how you can apply binary models to multiclass problems. ___ I hope you've found this Tweet helpful. Like/Retweet for support and follow @levikul09 for more Data Science content. Thanks 😉