Are we allowed to transform the continuous target variable by creating a log transformation in order to have a normal distribution?

Question

The following code gives the target variable Item_Outlet_Sales before transformation and Item_Outlet_Sales_log which is transformed

#treat extreme values in Item_Outlet_Sales
train['Item_Outlet_Sales_log'] = np.log(train.Item_Outlet_Sales)
test['Item_Outlet_Sales_log'] = np.log(test.Item_Outlet_Sales)

plt.figure(1)
plt.subplot(121)
sns.distplot(train.Item_Outlet_Sales)
sns.distplot(test.Item_Outlet_Sales);
plt.subplot(122)
sns.distplot(train.Item_Outlet_Sales_log)
sns.distplot(test.Item_Outlet_Sales_log);

Then use the new target variable (Outlet_Item_Sales):

#creating dummies for the training dataset
X = train.drop('Item_Outlet_Sales', 1) #drop the log target column
y = train.Item_Outlet_Sales_log

X = pd.get_dummies(X)
train = pd.get_dummies(temp_train)

Answer 1

If your objective is to convert non-normal data into something that looks more normal / gaussian - try the Box-Cox Transform here.

It is a family of transforms that looks at your data - and provides the best possible transformation.