Top Tweets for #LabelEncoding

One-Hot encoding vs Label encoding Both techniques turn categorical values into numbers. But what is the difference then? Let's discuss 👇 Most ML algorithms will struggle with categorical data. To avoid this, we usually use One-Hot encoding or Label encoding. After the transformation process, we can train the models on numbers. One-Hot encoding This technique creates a new feature for every unique categorical value. If we have a dataset with 3 colors, one hot encoding will create a new dataset with 3 new features. That can lead to issues as well because for too many categories the dimensionality will increase rapidly. For that reason, One-Hot encoding is better for data, where the number of categories is not large. Note: By default One-Hot encoding usually uses K dummies for K categories. But that is not effective and can lead to issues. K-1 variable is enough, but more on this in another post. Label encoding This technique replaces each unique categorical value with a consecutive number. For the same example dataset we will not have 3 new features, only 1. So computationally it is more effective, but it still has drawbacks. For example, the consecutive numbers can lead to a false impression about ranks between the values. If Red is two and Green is 1, one could interpret it as Red > Green. So which encoding technique to use? It depends on the dataset or the model you want to use. Use One-Hot encoding for not ordinal categories and less features. Use Label encoding with ordinal data, or where the number of categories is large.