How to predict calculation based data for Machine Learning?

Question

I have a large set of sample data as following:

The Price depends on the values from Val_1 to Val_14 (total 150 features). I want to use this table as a learning data set for Machine Learning. Now what I really want to implement is to predict the price based upon these sample data.The values can be altered and the predicted price should vary based upon these calculation.

I don't have any specific calculation formula for the price calculation so I want to use these dataset as a learning data for data classifier in Machine Learning.

I have tried implementing various algorithm based on Deeplearning using Theano with Python but upto now I have only succeed in predicting output having predefined output classes. Is there any method that uses Machine Learning algorithm to predict these kinds of calculation?

Answer 1

I am not sure deep learning is the best approach in this case - you have a very limited set of features and predicting a single value. I have a feeling that even a simple regression will give you acceptable values.

My suggestion is that you try to play with the different models that H2O offers. In addition to having many model types built-in it will also plot various metrics to help you choose what best works in your case. In my experience it is almost always GBM that gives the best results.

When you find what you are looking for, you can export the model and directly use it in java. Alternatively most model types have equivalent in python's sklearn .

Answer 2

There are tons of regression algorithms that you could use. Check the sklearn doc from python and read about the models and which one best fits to your specific case.

The linear model is a very simple one, but extremely sensitive when working with messed data.

I suggest you try the SVR from SVM.