Principle:DistrictDataLabs Yellowbrick Missing Data Visualization
| Knowledge Sources | |
|---|---|
| Domains | Data_Quality, Visualization |
| Last Updated | 2026-02-08 05:00 GMT |
Overview
Technique for visually diagnosing the extent, pattern, and distribution of missing values in a dataset to inform imputation and data cleaning strategies.
Description
Missing data visualization reveals the structure of missingness: whether it is random (MCAR), conditionally random (MAR), or systematic (MNAR). Bar charts show the count of missing values per feature, while dispersion plots show the exact sample indices where values are missing. Class-stratified views reveal whether missingness is correlated with the target variable, which has implications for model fairness and accuracy.
Usage
Use this principle during the initial data quality assessment phase, before imputation or feature selection, to understand missingness patterns and choose appropriate handling strategies.
Theoretical Basis
Missing data mechanisms (Rubin, 1976):
- MCAR (Missing Completely at Random):
- MAR (Missing at Random):
- MNAR (Missing Not at Random): Missingness depends on the missing values themselves
Pseudo-code Logic:
# Abstract algorithm
for each feature:
count_missing = sum(is_nan(feature))
positions = indices_where(is_nan(feature))
plot_bar(feature_name, count_missing)
plot_scatter(feature_name, positions)