Automatic Expert Advisor creation using WEKA and MT5 - how to improve the idea

[PhoinixY]

Hello, I tried using WEKA together with MT5 to automatically create an Expert Advisor based on the WEKA model.
I've created the following workflow:
1. Select Data: In MT5, select the data to be used to train the model. Only the default Indicators and price data can be selected.
2. Select the WEKA Classifier: Only the default WEKA Classifiers can be used.
3. Run the Strategy optimizer: This will train the model in WEKA.
4. Evaluate in MT5.

So a WEKA model is trained with the selected data. Through the strategy tester, a variety of combinations are evaluated, the model with the highest profit is then selected and saved.

Unsurprisingly, back testing looks fantastic, but forward testing looks bad (sometimes it is decent, but never good). You see it in the picture, Forward testing starts at the red line in the demo.png image below(can't insert pictures).

This example is on EURCHF-H4, and it uses current price data (open, close, high, low), two data points from AMA, two data points from RSI, and volume data. The optimizer will select how far in the past these data points are. This uses the Multilayer Perceptron with the default settings.

The process is slow and there is no money management yet.
What I'm hoping for here is:
1. What kind of data combinations or classifier would be a good idea. More data, fewer data?
You might say, simply try it, but a single try takes hours.
2. How to fight overfitting in this particular case.
3. Whether this approach has potential in your opinion.

Thanks for reading.