## The Missing Data Problem
Real sports datasets have missing values: matches with no xG data, player statistics for injured players, weather data gaps. How you handle missing data significantly affects model quality.
## Types of Missing Data
**MCAR (Missing Completely At Random):**
The probability of data being missing is unrelated to the true value or any other variable. Example: a random server error that dropped some match data. Safe to delete these observations.
**MAR (Missing At Random):**
The probability of data being missing depends on other observed variables but not on the missing value itself. Example: xG data missing for lower leagues (smaller league → less data collection). Can be imputed using observed data.
**MNAR (Missing Not At Random):**
The probability of data being missing depends on the missing value itself. Example: match attendance data missing when attendance was unusually low (embarrassing for the club). The missingness is informative — imputation is problematic.
## Imputation Strategies
**Mean/median imputation:** Replace missing values with the mean (continuous) or mode (categorical) of the observed values. Simple but ignores patterns.
**Regression imputation:** Model the missing variable as a function of other variables and use the predicted value. Better than mean imputation.
**Multiple imputation:** Create multiple datasets with different imputed values, run the model on each, and combine results. The gold standard for MCAR/MAR situations.
**Indicator approach:** Add a binary "is_missing" feature alongside the imputed value. Lets the model learn whether missingness itself is predictive.
## The Practical Recommendation
For betting models: track the volume of missing data per variable. If > 20% of observations for a key feature are missing, the feature may not be reliable enough to use. Below 5%: any imputation method works adequately.
Create a free account to track your progress and save bookmarks.
