I'm working on a first project in machine learning, with the aim of forecasting foreign exchange rates.
I am not trying to be specific about the currency, merely I am interested in what the historical price pattern can tell us about the price in the near term.
I have gathered historical prices for several different currencies, but I am wondering about how I should preprocess the data before training. I feel like it might be best to just convert the time series of prices to a time series of %change in price (by just subtracting the price at time $t$ from the price at time $t+1$, and then normalizing this appropriately), but I don't know if that would truly help to 'normalize' the data.
Can anyone offer some insight?
