How to approach deep learning classification with multiple inputs and a trinary state output applied to stock trading?

Hi,

I wonder what would be your approach to this classification problem that applies to stock trading.

Let's say I receive trending stock alerts form 10 sources at different times during the trading session for different stock prices. I have two risk management strategies to buy a given stock that produced different gain performances and that from statistics depend on the combination of AlertSource+TimeOfDay+Price.

So depending on the performance I got with my two risk management (RM) strategies I would like the algo to tell me which action I should take:

Buy with RM1=1 (if RM1 would give a better return than RM2)

Buy with RM2=2 (if RM2 would give a better return than RM1)

Don't buy=0 (if both RM1 and RM2 would result in a loss)

So it seems like a 3-input situation with one trinary state output (0, 1, 2) that depends on past values of two parameters (PerfRM1 and PerfRM2).

Thanks for your advice!

F

File example:

AlertSource TimeOfDay Price($) PerfRM1(%) PerfRM2(%) Action

1 09:35 15.95 0.5 0.9 2

4 10:08 26.32 1.2 0.7 1

6 10:43 9.97 -0.2 0.3 2

3 11:06 30.15 0.6 -0.1 1

2 12:09 18.99 -0.4 -0.6 0

. . . . . .

. . . . . .