I am working on Azure ML Studio and try to create a regression model to predict a numerical value. I will try to describe my features and what I have done until now.

My data with about 3 million rows :


  • 8 integer features from 1 to 25
  • 2 boolean features with 0 and 1
  • 3 integer features from 1 to 10
  • 2 integer feature from 0 to 500.000 (and 1.000.000 respectively) with about 4.500 unique values
  • 1 integer feature from 20 to 50
  • 1 integer feature from 1 to 15
  • 1 integer feature from 0 to 100


  • Integer from 10.000 to 100.000.000 with about 5.000 unique values

What I have done:

  • Split the dataset to 80% (train) and 20% (test). Then I split the training dataset again to 60% (actual train) and 40% (validation).
  • Normalize the features with many unique values (4th bullet in the above list)
  • Train a model of Boosted Decision Tree Regression.
  • Use the Sweep Parameters module to find the best combination

I tried also Neural Networks, Bayesian Linear Regression, but BDTR gave the best score.

I tried to exclude columns and start with only a few (based on what I think it will affect the model) and then add more columns one by one.

However, the least MSE I could achieved was 1.500.000 (plus I had many negative scored values)

So, I was thinking what other techniques I could use to improve the model.

  • $\begingroup$ If 10 is the min value the response can take on, I would say any prediction below 10 should be set to equal 10. You can also look into random forest/bagging, or possibly taking the average prediction of many Boosted trees/Neural net models to see if it helps your results a little. Also, another loss function might be nice (estimate 11 and truth 10 is a 10% error, but only loss 1, where estimate 1,100,000 and truth 1,000,000 is still 10% error, but the loss is 100,000 so it is prioritizing those higher values. Just something to consider, a lot depends on the context of the variables. $\endgroup$
    – TBSRounder
    Dec 24, 2015 at 17:44

3 Answers 3


I agree with @Hoap. Your features might be low for the amount of training observations you have. Instead of excluding columns, see if you're missing more features. Feature Engineering, rather than Feature Selection.
However, if you are looking for Feature Selection, then Azure ML has a Feature Selection Module with the option to specify how many features you'd like to keep.

Some simple verifications to do before you jump into modeling:

  • Visualize your dataset for any non-linear relationships.
  • You could also perform a simple correlation analysis to check for multi-collinearity.
  • I also think that normalizing all of your data between 0 to 1 for consistent comparable values between features would be helpful.

Hopefully one of these will show some unexpected pattern in your data. I apologize if you've already performed these checks. Just wanted to put them out there.

Looks like you pretty much used every regression model in the Azure ML library.


I think the next option you have to take is to add more features. You have a huge amount of training examples, which is good, but the number of features is very low. Adding more features is one of the most used methods to improve performance in machine learning.

Furthermore, it would be good to try to understand how your features affect your model. Imagine you have a linear model like y = theta1*feat1 + theta2*feat2 + theta3*feat3. If theta3 is near to 0, then feat3 is not affecting the model.


The best way to improve is doing error analysis, which means going through the errors the model makes in validation trials and seeing how we can improve, either by adding terms interactions, polynomials (to model nonlinear relationships), or simply using other models your data might be non-linear too. Also, try doing more feature engineering, cleaning, and preprocessing to best optimize model performance.


Your Answer

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

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