Hi Alejandro!

Thanks for the feedback :)

About the labels. I think it makes more sense to do it individually. So, based on recent price activity, is a particular trade on a currency profitable or not. Yet, if you're making the decision based on the pair of assets, maybe it makes sense to look at both jointly. It really depends on your goal.

For features, here's a possibly interesting article for you: https://medium.com/towards-data-science/machine-learning-for-forecasting-transformations-and-feature-extraction-bbbea9de0ac2

You should also take a look at Python's ta-lib for technical analysis