I use apply to loop the rows and get the column names of feat1, feat2 or feat3 if they are equal to 1 and scored is equal to 0. The column names are then inserted into a new feature called reason.

This solution doesn’t scale to larger dataset. I’m looking for faster approach. How can I do that?

df = pd.DataFrame({'ID':[1,2,3],
             'feat1_tax':[1,0,0],
             'feat2_move':[1,0,0],
             'feat3_coffee': [0,1,0],
             'scored':[0,0,1]})

def get_not_scored_reason(row):
    exclusions_list = [col for col in df.columns if col.startswith('feat')]
    reasons = [col for col in exclusions_list if row[col] == 1]
    return ', '.join(reasons) if reasons else None

df['reason'] = df.apply(lambda row: get_not_scored_reason(row) if row['scored'] == 0 else None, axis=1)

print(df)

   ID  feat1_tax  feat2_move  feat3_coffee  scored      reason
0   1          1           1             0       0  feat1_tax, feat2_move
1   2          0           0             1       0           feat3_coffee
2   3          0           0             0       1                   None