In our previous articles, we discussed mean or median imputation, end-of-distribution imputation, and frequent category imputation. But, sometimes we may want to fill in missing numerical values with numbers that indicate missing values. For example, if we are imputing missing values in the age column, we can impute the number 9999 or -1 in the place of missing numbers. As a result, the distribution of data will not be impacted, yet missing numbers will be handled.

To summarize, in arbitrary value imputation, we replace missing values in a numerical column with numbers that are chosen arbitrarily. And we select the arbitrary numbers in such a way that the numbers do not belong to the dataset. The numbers rather signify that there is a missing number in the column.

Let’s read the titanic dataset and see the percentage of missing values in each column of the dataset.

import seaborn

df = seaborn.load_dataset("titanic")
print(df.isnull().mean()*100)

As we discussed previously, print(df.isnull().mean()*100) will print the percentage of missing values in each column of the dataset. The output will be like the following:

survived        0.000000
pclass          0.000000
sex             0.000000
age            19.865320
sibsp           0.000000
parch           0.000000
fare            0.000000
embarked        0.224467
class           0.000000
who             0.000000
adult_male      0.000000
deck           77.216611
embark_town     0.224467
alive           0.000000
alone           0.000000
dtype: float64

So, the age column of the dataset has 19.865320% missing data. Let’s fill in the missing ages with the number -1. We can use the following Python code for that purpose: