<aside> 💡 Code examples in Python Feature Engineering Cookbook

</aside>

Equal-Frequency Binning

In equal-frequency binning, we sort the data values of a continuous variable into bins that contain the same number of observations. The quantiles are used to determine the bin edges. The resulting intervals may not have equal width, and that’s OK.

For example, if we want to sort the variable income into 5 intervals of equal frequency, we would determine the 20th, 40th, 60th, 80th, and 100th quantiles to find out the limits of the bins.

The beauty of equal-frequency binning is that it improves the value spread of skewed variables.

Advantages and Disadvantages

Equal-frequency binning has some advantages and disadvantages, which I describe below:

Advantages

• It is a good option when the distribution of data is skewed and you want to ensure that each bin has a roughly equal representation of the data.
• It is suitable for data that is not normally distributed or contains outliers.

Disadvantages

• The number of data points in each bin may not be equal due to rounding errors or an uneven distribution of the data.
• If the variable is very skewed, we may need to chose a small number of quantiles, otherwise some bins may end up empty.
• It may result in a loss of information by putting in the same interval observations that are associated with a different (target) class.

As with equal-width discretization, in equal-frequency, the number of bins is determined arbitrarily by the user and its value might need to be optimize to extract the maximum value from a variable.