Binary Classification vs Regression about crypto_trader

[bshaw2019]

I definitely agree that a regression model is more appropriate for projected price/profit.

The main function of the classifier is generating a confidence value that the price will increase or decrease using a confusion matrix.

Perhaps we should split the trading pair scoring process into two separate models, a regressor and a classifier. These could be used to generate predictions for both reward (projected profit from the regressor) and risk (the confidence value for a price increase or decrease move). I think building these models for a variety of timeframes for all symbols (example: 1 min price change, 5 min, 15 min, 30 min, 1 hour) will provide the Q-learning trader with enough information to determine an optimized trading strategy that both reduces risk and rewards profit.

[andrewvaliente]

regression for prices, classification for tweets