0
$\begingroup$

I have a dataset with the following format:

  • Rows: 3700,
  • float_columns: 17,
  • int_columns: 2,
  • categorical_columns: 12
  • Target Type: Continous, float

My dataset is an insurance dataset that stores the losses associated with claims. I would like to build a regression model that predicts the losses based on the input features.

I'm thinking of using lightGBM's LGBMRegressor but I don't know how good this is at general regression tasks. What are the methods and models that are typically used in the industry?

$\endgroup$

1 Answer 1

1
$\begingroup$

The performance of any model is data-dependent, so the exact empirical performance is unknown until the model is trained. In general, for regression tasks, neural networks may perform slightly better because of the rigidity of tree models compared to neural networks.

I think depending on how much computational power you have available, most of the models are quite lightweight, so I would try all of them; all of the models are used in different contexts.

$\endgroup$
2
  • $\begingroup$ How does the training time of a tree and deep learning model compare on a small dataset? $\endgroup$
    – Connor
    Feb 8 at 14:47
  • $\begingroup$ To compare the time of the different types of models, you can sort of sketch the amount of computations that are needed to complete training. I would expect the amount of operations for trees to be less than deep learning because gradient descent requires a fair amount of operations. However, they can be optimized on GPUs and such so could be faster, I don't know you would need to test to find out exactly. $\endgroup$
    – timmy1691
    Feb 8 at 19:54

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.