I have a dataset with both missing values and outliers in continuous features. I would like to perform Box-Cox transformation on every continuous feature to reach the best distribution. Box-Cox works only on positive data. When I fill missing values with mean before removing outliers the outliers affect the mean values. When I first transform data with Box-Cox with dropna() function and after that remove outliers and fill missing values it probably would affect the distribution of data so what would be the best workflow here?