BelkaMiner is a fully automated breakout/momentum/reversal trading system (platform) based on Neural Networks and the most well-known clustering algorithm that is used for unsupervised Machine Learning and statistical data analysis (Data Mining).
- The EA allows you to build a machine learning-based trading system on any pair and timeframe;
- It's a real Data Miner and ML system, not just pretty words. You can fully control the learning process and change the data mining settings in your own way;
- The system can learn how to trade breakout, momentum and reversal strategies by recognizing patterns in historical data;
- The EA groups a set of entry points in such a way that points in the same group (cluster) are more similar to each other than to those in other groups.
Then we test each cluster and try to find profitable ones with good performance;
- The Bollinger Bands, Donchian Channel, Price Action and 6 low correlated Input variables with the Outlier data filter are available;
- You can share ML-based strategies with the community, the Data Mining results are stored in a text file;
- The system does not use Martingale, grid, hedging and other risky Money Management techniques;
- All settings are fully customizable, no hidden parameters;
- This is a long-term project that will last at least 10 years. The EA is based on Belka Core, so that both EAs will be developed together.
Please read this post first: How to Use BelkaMiner. Visit my website for more info
ML-based strategies may be sensitive to brokers, quotes, and historical data!
I would say the current release is still in beta phase, new ML methods, techniques and strategies/set-files will be added in the future. So DO NOT buy this EA if you want easy and quick money. Buy it if you want to learn more about ML technologies and use them in your trading
Default settings of the EA contains a mean reversion strategy that was created using clustering on the EURUSD pair and the M5 timeframe. The Default_EURUSD_M5 settings file will be created automatically after the first run
Strategies and set-files can be found on the settings page, as well as on my website
- ML-based systems are called Black-box to indicate that their function is obscured. In terms of ML, it is very difficult to figure out what has been learned;
- Therefore, there is a high risk of over-fitting. That's why I do not use ML-based models to generate trading signals; I only use them to analyze and classify signals generated by the simple White-box algorithm. In other words, I use Gray-box models. This allows the system to be more likely to pass the out-of-sample test at the first run;
- In ML-based trading, the way the result is obtained (the process of strategy creation) is more important than the result;
- The EA is not a trading strategy; this is a system (platform) for creating ML-based strategies. So the platform cannot trade in profit, loss, or zero; it's just a set of tools containing models, filters, etc. A trading strategy is a set of predefined rules contained in a set-file telling the EA when to buy and sell. Therefore, please do not write that the EA trades bad or good; the system follows rules of a loaded trading strategy, the main goal of the platform (EA) is running the strategy with no errors and nothing else.
- You can use the demo version for creating strategies and educational purposes.
- Account type: any;
- Account balance: any;
- Fast PC;
- Brain and desire to study ML-based models together with me.
- I strongly recommend using the MT5 version for data mining and backtesting. It works much faster. It's enough a demo version, not necessary to buy it. You can use SetFileConverter from my website (Download section) to open/view a set-file or convert it to the MT4/MT5 version;
- It is recommended splitting historical data into an in-sample period used for the initial parameter estimation, data mining, and an out-of-sample period used to evaluate forecasting performance. This helps to slightly reduce the risk of curve-fitting. But remember, out-of-sample testing is not the panacea;
- It is recommended using historical data from different providers/brokers.