0
$\begingroup$

I want to predict price of used cars. I have data like this:

Data

Is this problem suitable for deeplearning or Should I use XGBOOST, RandomForest etc.?

I used one hot approach for nominal features and scaled numeric values between 0-1. Is there anything I can do to improve my data?

I used deeplearning4j on java and predict prices with 10-15% error but I want to predict prices with 1-2% error

$\endgroup$
1
  • 1
    $\begingroup$ Be careful: ML approaches might give you a 1-2% error margin on average, but you can't guarantee that every individual prediction is within a 1-2% error margin. This means that you might find cases where the predicted price is 200% off the real value. $\endgroup$
    – Erwan
    Jun 20 '19 at 12:20
0
$\begingroup$

The problem is a case for "deep learning", I think, and I would start with boosting. I have good experiance with catboost (in Python). It can handle factors very well ("catfeatures", encoded as dummies) and it is relatively fast. Also lightgbm might be a good option. Also xgboost is an option.

I think data-wise one hot encoding and scaling is the first thing to do. Things beyond that is a bit try and error. The most important part might be that you tune your hyperparameter well (during boosting).

Catboost (Python): https://catboost.ai/docs/concepts/python-usages-examples.html LightGBM (Python): https://lightgbm.readthedocs.io/en/latest/Python-Intro.html

Happy coding!

$\endgroup$
0
$\begingroup$

I would say start with data cleaning to get a better sight of the problem. Remove all the unnecessary fields, this way you can get a clear picture of which algorithm to use. If no of categorical features are higher than non-categorical features in your data, Trees will work better

$\endgroup$
0
$\begingroup$

If you don't need uncertainty estimates around your predictions, I would recommend just using XGBoost regression and call it a day.

$\endgroup$

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.