I am new to machine learning, so was hoping you might have some insights one what classification algorithm might be best suited for my data.

I have a data set of hundreds of predictions from each of ~100 forecasters. Each prediction is binary, and the outcome variable is binary.

So basically let's say I have data on 200 events. For each event, I have 1 prediction from each of 100 forecasters. Each of the 100 forecasters predicts each of the 200 events.

I am hoping for an output of some formula, for which I input the predictions of the forecasters for that particular outcome, and it renders an output for the probability of that outcome.

My main interest is in predictive accuracy.

Do you have any suggestions for what algorithms might suit this problem?

  • $\begingroup$ did you do any analysis on your data? anything like looking at how often the forecasters get things right? how often is their average off etc.? do you expect any dependencies between forecasters or events? $\endgroup$
    – oW_
    Commented Aug 24, 2017 at 19:36
  • $\begingroup$ I have looked at it some. They are correct roughly 60% of the time. The average is correct perhaps 70%. Forecasters are correlated with each other, and some are higher correlated with each other than others. $\endgroup$
    – evt
    Commented Aug 25, 2017 at 19:13

1 Answer 1


I would say support vector machine, support vector classifier. This will allow you to have as many predictors as you want. Also, you can improve your accuracy by tuning your model parameters.

Here is another nice link for support vector classification

Since you did not mention any language, I would use python as there are so much online support and you resources available online.

I would also check xgboost classifier.

Let us know how did it go.


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.