Using Data classification (Machine Learning) to assist trading

Hi,

I have been following this site for a VERY long time and I have used many strategies published here. From OzFX, Cowabunga, 3 ducks and many other.

I am a professional software developer by nature so it is in my line of work to back test, automate and try out various strategies to see how effective they are. I stumbled onto the field of machine learning and started investigating its application to the forex field. I initially developed a very basic version of an application (using FXCM as data source) that ran on my pc but have subsequently published it to a website to make it publicly accessible.

I am currently using this info as trend confirmation. Over the past month I have had a 74% accuracy with trend. PLEASE NOTE this is trend confirmation, NOT price confirmation. This is where I think lots of people give up on machine learning. They want accurate price predictions, this is impossible.

So, my current strategy looks at trend (like I explained above) and then I do manual trades based on other strategies that I found on this site. I found that it really pushes up my success rate when doing it this way. So effectively machine learning is only (in my case) an additional trend confirmation indicator in my bag of indicators.

For those interested, I am currently only analyzing 4 currency pairs (EUR/USD, USD/GBP, USD/CHF and EUR/GBP) here is a screenshot of one of the pics on the site.


At the moment I feel what is lacking is market sentiment. I have developed another system that uses binary classification to determine market sentiment based on twitter feeds. I will publish it in the next month or so (things are a bit hectic at work at the moment)

Please let me know if you are interested, then I will try and do it sooner.

Regards
Charl

Been a while since my last post but things where kinda hectic, luckily it will start slowing down for the festive season.

I have done lots of testing in November as well as publish a website with the live predicted hourly/daily charts for 4 currencies (EUR/USD, USD/CHF, EUR/JPY, GBP/USD). In the live testing I have done I have noticed that the GBP/USD and USD/CHF are more accurate then the USD/CHF and EUR/JPY (specifically on the hourly charts). I can replace them at a later stage with different pairs or just add more if needed. For now I first want to get it working as good as possible and then I will add more currency pairs.

I have also created an application that automatically trades all these signals on an demo FXCM account. It is a bit buggy still but I will be running it on a demo account and will keep you posted once I have a working version.

If you are interested please have a look at it, the site is located at Trademotions.com - Using machine learning to predict market movement. I would like to make it a community project if possible so any feedback would be greatly appreciated to help try and optimize it.

Oh, I am also adding an api so you can directly get to the predicted/actual data if you like. I will expose share this at a later stage for those interested.

Lastly (and very important) : The chart displays all dates/times in GMT (that is how I get it from FXCM) so if the current GMT time is 5:45am the first predicted candle time will be 5:00am because the actual 5:00am candle is not closed yet you will see the predicted closing price for that time. I hope this makes sense if not please give me a shout (or a suggestions on how to make it less confusing :slight_smile:

What are you using for trend confirmation?

Hi,

For the last two month I found that the best combination for me is using only the hourly predicted direction (not the predicted price but mainly just the direction, so if price is lower than the previous price, I only do short trades and if the next predicted price is higher I only do long trades) This is currently my trend confirmation. Most of my trades I close within the hour.

Hope this helps, please shout if you need more info etc.

Regards