Has anyone here tried to predict a commodity's price by using other commodities prices as features in a machine learning algorithm? What techniques have been successful?


1 Answer 1


Based on your question there are couple of things which I would assume to answer your question:

  1. As you need to predict the commodity price the data which is collected is time series data.
  2. Since you want to use other commodity to predict, it means that you don't have any past data of the product which you want to predict.

The answer could be derived by performing some exploratory analysis on the existing data. i.e., based on your business understanding you need to decide which product is similar to the new product. This kind of techniques is used to understand the sale of the new product/how is it going to perform after the launch.

Techniques which can be used over here are Time Series Analysis like ARIMA, if seasonality is present then SARIMA, if no trend then Exponential Smoothing(too many spikes), there are other models like Auto Regression, Moving Average, Croston if there are 0's etc.

This is one way of looking at your problem.

  • $\begingroup$ Thanks for the answer. I am trying to predict bent crude price for the next month with the data avaliable upto the current month for bent crude as well as several other commodities. So far linear regression is giving me the best results. I am able to accurately predict which direction the price will move (up or down), but the price prediction still has error of an average 15% to 20%. Has anyone been able to achieve better performance? $\endgroup$
    – Vin
    Dec 6, 2017 at 1:52
  • $\begingroup$ Did you do more feature engg, is there any way where you can get some other factors which are effecting your target variable? 15 to 20%, I won't say that your model is the best(very subjective to your project) but do if possible spend abit more time on feature engg(deriving new variables, removing outliers etc), by doing so you can improve your prediction accuracy. Other techniques to improve accuracy is (Bagging and Boosting). Why did you choose regression over time series analysis? $\endgroup$
    – Toros91
    Dec 6, 2017 at 2:04

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.