Addressing missing data

Analisi esplorativa dei dati in Python

George Boorman

Curriculum Manager, DataCamp

Why is missing data a problem?

Affects distributions
- Missing heights of taller students
Less representative of the population
- Certain groups disproportionately represented, e.g., lacking data on oldest students
Can result in drawing incorrect conclusions

Sample versus population height distribution, showing the sample has a lower maximum and mean

Data professionals' job data

Column	Description	Data type
`Working_Year`	Year the data was obtained	Float
`Designation`	Job title	String
`Experience`	Experience level e.g., `"Mid"`, `"Senior"`	String
`Employment_Status`	Type of employment contract e.g., `"FT"`, `"PT"`	String
`Employee_Location`	Country of employment	String
`Company_Size`	Labels for company size e.g., `"S"`, `"M"`, `"L"`	String
`Remote_Working_Ratio`	Percentage of time working remotely	Integer
`Salary_USD`	Salary in US dollars	Float

Salary by experience level

Box plot of salaries by experience level using the clean dataset, with an upper limit near 600000 dollars

Box plot of salaries by experience level using a dataset with missing values, showing an upper limit near 450000 dollars

Checking for missing values

print(salaries.isna().sum())

Working_Year            12
Designation             27
Experience              33
Employment_Status       31
Employee_Location       28
Company_Size            40
Remote_Working_Ratio    24
Salary_USD              60
dtype: int64

Strategies for addressing missing data

Drop missing values
- 5% or less of total values
Impute mean, median, mode
- Depends on distribution and context
Impute by sub-group
- Different experience levels have different median salary

Dropping missing values

threshold = len(salaries) * 0.05
print(threshold)

Dropping missing values

cols_to_drop = salaries.columns[salaries.isna().sum() <= threshold]

print(cols_to_drop)

Index(['Working_Year', 'Designation', 'Employee_Location',
       'Remote_Working_Ratio'],
      dtype='object')

salaries.dropna(subset=cols_to_drop, inplace=True)

Imputing a summary statistic

cols_with_missing_values = salaries.columns[salaries.isna().sum() > 0]
print(cols_with_missing_values)

Index(['Experience', 'Employment_Status', 'Company_Size', 'Salary_USD'], 
    dtype='object')

for col in cols_with_missing_values[:-1]:
    salaries[col].fillna(salaries[col].mode()[0])

Checking the remaining missing values

print(salaries.isna().sum())

Working_Year             0
Designation              0
Experience               0
Employment_Status        0
Employee_Location        0
Company_Size             0
Remote_Working_Ratio     0
Salary_USD              41

Imputing by sub-group

salaries_dict = salaries.groupby("Experience")["Salary_USD"].median().to_dict()

print(salaries_dict)

{'Entry': 55380.0, 'Executive': 135439.0, 'Mid': 74173.5, 'Senior': 128903.0}

Imputing by sub-group

salaries["Salary_USD"] = salaries["Salary_USD"].fillna(salaries["Experience"].map(salaries_dict))

No more missing values!

print(salaries.isna().sum())

Working_Year            0
Designation             0
Experience              0
Employment_Status       0
Employee_Location       0
Company_Size            0
Remote_Working_Ratio    0
Salary_USD              0
dtype: int64

Let's practice!

Analisi esplorativa dei dati in Python