Introduction to Machine Learning on a Forex Trading Project

Machine learning is integrated in multiple domains, it permits to a program to learn and optimize its execution without explicitly programming it.

TOM MITCHELL gives a good definition «A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P, if its performance at tasks in T, as measured with P, improves with Experience E ».

Machine Learning is also applicable to Forex Trading and I'm beginning a project where I will implement it from scratch to build a EA (Expert Adviser) for Metatrader4 and share with you articles of my studies and progression, In this project as we will deal with Forex trading system we can define Experience Tasks and Performance this way:

• EXPERIENCE: the experience of making trades.
• TASKS: the tasks of opening and closing orders.
• PERFORMANCE: making a profitable trade.

The supervised learning is used when we have a set of data, and we already have an idea about the correlation between the inputs and the outputs, we can recognize two types of it, regression and classification.

The best example of a regression situation is when we should predict the price of a house knowing its size, as we know that the price of a house grows as the size of the house grows, if we have some data entry about current price of some houses and their size we can predict the range of price of a house as we know it size.

Classification is just like regression but the output values are discrete or categorized, as an example we can say knowing the person face characteristics we should predict either is it a male or female.

In the regression example, the output is a price so the output is not countable in the classification example we have only two values male female, our supervised learning studies applied to Forex trading we will apply classification as the output will be profitable/ not profitable trade or long / short trade.

Applying machine learning on Forex trading is not a simple task as we must define the case properly, as we will begin from scratch let’s first define what the goal of creating this program, the goal must be measurable, reachable, and of course have sense.

Let’s say the program must predict the price of an asset, really, this is the dream of every trader but this definition is very general, not measurable and must be decomposed into sub non-contradictory goals, if the price climbs 1 pip then goes down should we consider this 1 pip rising or, must we consider 10 – 100 pips, to be more realistic let’s rather say that the program must open on profitable order with a predefined take profit pips and stop loss pips.

PERFORMANCE P: Opening a profitable trade (Long/Short) with predefined (Take Profit/Stop Loss) pips.

The next step consists on defining some correlated properties that influence the price trending on market this properties must be also measurable, for that we will test some Forex indicators there is an uncountable indicators ready to use on different platforms with an uncountable ways to parameter them we will study some very known indicators as Relative Strength Index, Stochastic, MACD, Moving Average, Parabolic Stop And Reverse, Fractals, Fibonacci, and will add a simplest version of the position of the price in a predefined last floating period, I like this indicator that we will explain later.