Artificial Intelligence EA ( by Yury V. Reshetov )

Hi all,

This thread is for discussion of Artificial Intelligence EA by Yury V. Reshetov

The description below is from it’s MT4Appstore page:

Artificial Intelligence Forex expert advisor uses a very simplistic imitation of the neural network to produce buy and sell signals and trailing stop losses. It is not a real neural network, because it doesn’t learn from market, instead you need to optimize it to the market to set the most fitting parameters. Its perception function uses the Bill Williams’ Accelerator/Decelerator oscillators, which are weighted according to the set parameters. Expert advisor also performs checks for available free margin to stop trading if it is bankrupt.

The original EA may be found here: Artificial Intellegence - MQL4 Code Base

I hope to engage in lively discussion with other users of this interesting EA and share optimization tips, tweaks & improvements.

Thanks and regards,
Dudest

Attached are latest MT4 strategy test results for GBPUSD H1 and GBPJPY H1 for Jan 2nd - June 8th 2012.

Below are optimized input parameters suggested by strategy testing. In all, the SL=40 and LOTS=0.01. The optimized parameter is PROFIT FACTOR (pf)

If anyone has more test results, please share, would be great to compare!

PS: i plan to be doing optimization weekly ( testing on H1 for 2-month timeframe ) and adjust the parameters in the EA accordingly.

======================

GBPJPY: pf 9.23

x1 = 200/200/200
x2 = 110/110/110
x3 = 30/50/40
x4 = 30/40/30

==

x1 = 190/190
x2 = 120/110
x3 = 50/40
x4 = 40/40

======================

GBPUSD: pf 3.29

x1 = 100
x2 = 120
x3 = 10
x4 = 140

======================

AI_GBPUSD & GBPJPY_Strategy_Tests.zip (78.8 KB)

Here are some problems that I can see. With a neural network approach, the idea of a neural network is that the NN learns, over time, when the best time to do a trade and to exit a trade. You do have to train it. But the training and learning is done by the neural net. You tell it a time frame, start & finish for the training period.

With this artificial one, the user has to optimize it. What does that mean? When do you have to “retrain” it? What is optimization?

If you don’t use a long enough time period for training you may get lower returns or negative returns.

I don’t think this approach will work for most users.

It is not a neural network.

As the author states [I]“It uses Bill Williams’ Accelerator/Decelerator oscillators” [/I]

It would be better called [I]“Bill Williams’ Accelerator/Decelerator EA”[/I]

I think you could optimize it all day every day and come up with extraordinary results. Run a forward test and i suspect you will get completely different results. Your figures are a case of over-optimization & probably bad data.