I have an Advanced MT5 EA that trades correlated forex pairs. It has internal Virtual Testing mechanism and has the ability to choose the best pair and setting to trade.
I believe the process can be done more efficiently and correctly using ML. It could use MQL5 or other outside sources and apps. You can get Deeplearning, Keras or what not.
The program should work like this.
The program will run an optimization based on a given strategy and find a list of (1000 - 5000 or more) settings that have been profitable in the past i.e. (21 weeks, 9 weeks and 3 weeks). Pick 50 top settings and trade the best setting.
As it is trading, it must continously and nonestop optimise, find best settings, prioritize new list and have it ready to trade again.