1
$\begingroup$

I have past data of a large number of people who applied for a loan and their movement through 8 different stages, from start of application to loan being paid out.

I am trying to build a model that would predict how many people will get their loan completed (paid out) in the next 5 days (a working week). I'm using LightGBM and it would learn based on past data, and then use the model on people currently at various different stages and predict how many of them will likely be completed in the next 5 days. Those 8 stages can only happen one after another like a pipeline, so stages nearer to the last stage 'Completion' have a higher probability of being complete.

The model gives a good accuracy, but a poor precision score, there are too many false positives.

  1. I primarily used the Mean Squared Error, any recommendations for another evaluation metric?

  2. Since the 8 stages follow a pipeline in which they can only go one after another, can I somehow define them differently than I did?

$\endgroup$
  • $\begingroup$ Are you looking for a different loss function or a different evaluation metric? A good evaluation metric for imbalanced classification is F1-score. However, I bet you are looking for a loss function... $\endgroup$ – spec3 Jan 13 at 11:33
1
$\begingroup$

Regarding 2) Your task sounds like a classical case for survival analysis. From the description of your problem, I cannot really say if this would be an option, but I just want to let you know about it: https://en.m.wikipedia.org/wiki/Survival_analysis

$\endgroup$

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.