Log Transformations

As we have seen, Normalization and Standardization techniques are able to bring the different scale attributes to a common scale, but if the distribution is skewed, then it remains skewed after the scaling process. Log transformations can help in making a skewed distribution to a normal distribution or a highly skewed distribution to less skewed.

 

Few real world examples where Log transformations are done:

 

- Measuring Earthquake

- Measuring Sound

 

Now let us look at our last example where the attribute “Fare” is highly right skewed. 

 

Code:

import pandas as pd

import numpy as np

import matplotlib.pyplot as plt

import seaborn as sns

data['new_fare'] = data['Fare'].apply(lambda r: np.log(r))   # Apply log transformation

sns.displot(data['Fare'])

sns.displot(data['new_fare'])

 

Output:

 

As we can see that our original data was highly skewed and after transformation skewness is reduced. Now the transformed values are more visible.

 

Point to Remember:

• If you have negative or value as 0 in your distribution, then log transformation can’t be used.
• Log can only take positive values as input
• Log(base 10) and Log(base e), both can be used for log transformations. You should not see any accuracy related issues.