2
$\begingroup$

I have monthly data of loan installment repayment. The data contains basic features like salary, age, gender, credit score, etc. Along with the above features, I have the data for the last 6 installment failure/success. Now based on this, I want to predict which customers are going to default next month.

The problem I would like to highlight, to apply any machine learning algorithm, I need labels for the same month, i.e., for Feb 20 predictions, I need some rows with labels success/failure in Feb 20 (so that I can train model on this) and then can predict on the remainder of data from Feb 20.

But here I don't have any labels for Feb 20 (I have data for past failures instead).

$\endgroup$
1
$\begingroup$

This is a common problem. In a perfect world, training data is perfectly representative of the data that will need to be predicted. However, we don't live in a perfect world and we need to work with approximations. You don't have data about Feb-20 but you have data about Jan-20, Dec-19, etc. You can still use this data! Nothing is stopping you.

$\endgroup$
3
  • $\begingroup$ If i use Jan20 or Dec-19 then essentially i will be doing prediction for Jan20 by putting the failure/success from Jan only.. which is not right. $\endgroup$
    – SKB
    Jan 28 '20 at 13:08
  • $\begingroup$ No, you'll learn from Jan20, Dec19 and you will use what you learned to make predictions about Feb20, Mar20, etc. What are your input features? $\endgroup$ Jan 28 '20 at 13:15
  • $\begingroup$ That is what I want to understand how to use historical results to predict future(specifically I don't have any data for next month to train the model). If I train the model on Jan 20 data then prediction also can be done on jan data. The data contains basic features like salary,age,gender, credit score etc. Also last 6Months pass/fail of installment. $\endgroup$
    – SKB
    Jan 29 '20 at 3:20

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.