Articles on machine learning in trading

In this article, we will consider the algorithm that demonstrates the most controversial results of all those discussed previously - the differential evolution (DE) algorithm.

You might not have noticed that the matrix modeling was a little strange, since only columns were specified, not rows and columns. This looks very strange when reading the code that performs matrix factorizations. If you were expecting to see the rows and columns listed, you might get confused when trying to factorize. Moreover, this matrix modeling method is not the best. This is because when we model matrices in this way, we encounter some limitations that force us to use other methods or functions that would not be necessary if the modeling were done in a more appropriate way.

We invite you to get acquainted with the NAFS (Node-Adaptive Feature Smoothing) method, which is a non-parametric approach to creating node representations that does not require parameter training. NAFS extracts features of each node given its neighbors and then adaptively combines these features to form a final representation.

Tree-based classifiers are typically overconfident: true win rates near 0.55 appear as 0.65–0.80 and inflate position sizes and Kelly fractions. This article presents afml.calibration and CalibratorCV, which generate out-of-fold predictions via PurgedKFold and fit isotonic regression or Platt scaling. We define Brier score, ECE, and MCE, and show diagnostics that trace miscalibration into position sizes, realized P&L, and CPCV path Sharpe distributions to support leakage-free, correctly sized trading.

This article introduces a fairly effective method of multi-agent trajectory forecasting, which is able to adapt to various environmental conditions.

In this article, we will consider conformal predictions and the MAPIE library that implements them. This approach is one of the most modern ones in machine learning and allows us to focus on risk management for existing diverse machine learning models. Conformal predictions, by themselves, are not a way to find patterns in data. They only determine the degree of confidence of existing models in predicting specific examples and allow filtering for reliable predictions.

This article introduces the Conformer algorithm originally developed for the purpose of weather forecasting, which in terms of variability and capriciousness can be compared to financial markets. Conformer is a complex method. It combines the advantages of attention models and ordinary differential equations.

In previous articles, we discussed the Decision Transformer method and several algorithms derived from it. We experimented with different goal setting methods. During the experiments, we worked with various ways of setting goals. However, the model's study of the earlier passed trajectory always remained outside our attention. In this article. I want to introduce you to a method that fills this gap.

Category Theory is a diverse and expanding branch of Mathematics which as of yet is relatively uncovered in the MQL5 community. These series of articles look to introduce and examine some of its concepts with the overall goal of establishing an open library that provides insight while hopefully furthering the use of this remarkable field in Traders' strategy development.

In this discussion, we contrast the classical approach to time series cross-validation with modern alternatives that challenge its core assumptions. We expose key blind spots in the traditional method—especially its failure to account for evolving market conditions. To address these gaps, we introduce Effective Memory Cross-Validation (EMCV), a domain-aware approach that questions the long-held belief that more historical data always improves performance.

Soft Actor Critic is a Reinforcement Learning algorithm that utilizes 3 neural networks. An actor network and 2 critic networks. These machine learning models are paired in a master slave partnership where the critics are modelled to improve the forecast accuracy of the actor network. While also introducing ONNX in these series, we explore how these ideas could be put to test as a custom signal of a wizard assembled Expert Advisor.

We determine the overbought and oversold condition of the market according to chaos theory: integrating the principles of chaos theory, fractal geometry and neural networks to forecast financial markets. The study demonstrates the use of the Lyapunov exponent as a measure of market randomness and the dynamic adaptation of trading signals. The methodology includes an algorithm for generating fractal noise, hyperbolic tangent activation, and moment optimization.

In this article, I would like to introduce you to an interesting trajectory prediction method developed to solve problems in the field of autonomous vehicle movements. The authors of the method combined the best elements of various architectural solutions.

We follow up on our last article, where we introduced the indicator pair of the MACD and the OBV, by looking at how this pairing could be enhanced with Machine Learning. MACD and OBV are a trend and volume complimentary pairing. Our machine learning approach uses a convolution neural network that engages the Exponential kernel in sizing its kernels and channels, when fine-tuning the forecasts of this indicator pairing. As always, this is done in a custom signal class file that works with the MQL5 wizard to assemble an Expert Advisor.

Transformer architecture-based models demonstrate high efficiency, but their use is complicated by high resource costs both at the training stage and during operation. In this article, I propose to get acquainted with algorithms that allow to reduce memory usage of such models.

In the context of reinforcement learning, model procrastination can be caused by several reasons. The article considers some of the possible causes of model procrastination and methods for overcoming them.

Let's try mining CFTC data, downloading COT and TFF reports via Python, connecting all this with MetaTrader 5 quotes and an AI model, and get forecasts. What are COT reports in the Forex market? How to use COT and TFF reports for forecasting?

In this article, we discuss Directional Diffusion Models that exploit data-dependent anisotropic and directed noise in a forward diffusion process to capture meaningful graph representations.

We will add Deep Learning to those three examples that were published in previous articles and compare results with previous. The aim is to learn how to add DL to other EA.

In this article we will build a basic neuron. And although it looks simple, and many may consider this code completely trivial and meaningless, I want you to have fun studying this simple sketch of a neuron. Don't be afraid to modify the code, understanding it fully is the goal.

Here I will consider the fairly new Stochastic Marginal Actor-Critic (SMAC) algorithm, which allows building latent variable policies within the framework of entropy maximization.

We will analyze alternative data curated by the Chicago Board Of Options Exchange (CBOE) to improve the accuracy of our deep neural networks when forecasting the XAUEUR symbol.

The article considers the algorithm of the MEC family called the simple mind evolutionary computation algorithm (Simple MEC, SMEC). The algorithm is distinguished by the beauty of its idea and ease of implementation.

In the last piece, we concluded our look at the pairing of the gator oscillator and the accumulation/distribution oscillator when used in their typical setting of the raw signals they generate. These two indicators are complimentary as trend and volume indicators, respectively. We now follow up that piece, by examining the effect that supervised learning can have on enhancing some of the feature patterns we had reviewed. Our supervised learning approach is a CNN that engages with kernel regression and dot product similarity to size its kernels and channels. As always, we do this in a custom signal class file that works with the MQL5 wizard to assemble an Expert Advisor.

This article discusses the use of the Go-Explore algorithm over a long training period, since the random action selection strategy may not lead to a profitable pass as training time increases.

In this article, we will attempt to predict the market with a decent model for time series forecasting named DeepAR. A model that is a combination of deep neural networks and autoregressive properties found in models like ARIMA and Vector Autoregressive (VAR).

The Learning Rate, is a step size towards a training target in many machine learning algorithms’ training processes. We examine the impact its many schedules and formats can have on the performance of a Generative Adversarial Network, a type of neural network that we had examined in an earlier article.

In this article. we will discuss a method of complex multimodal interaction analysis and feature understanding.

Economic Calendar Data is not available for testing with Expert Advisors within Strategy Tester, by default. We look at how Databases could help in providing a work around this limitation. So, for this article we explore how SQLite databases can be used to archive Economic Calendar news such that wizard assembled Expert Advisors can use this to generate trade signals.

In today's discussion, we examine the strategy of multiple time-frame analysis to learn on which time frame our AI model performs best. Our analysis leads us to conclude that the Monthly and Hourly time-frames produce models with relatively low error rates on the EURUSD pair. We used this to our advantage and created a trading algorithm that makes AI predictions on the Monthly time frame, and executes its trades on the Hourly time frame.

Adaptation of the classical CAPM model for the Forex currency market in MQL5. The indicator calculates expected return and risk premium based on historical volatility. The indicators rise at peaks and bottoms, reflecting the fundamental principles of pricing. Practical application for counter-trend and trend-following strategies, taking into account the dynamics of the risk-reward ratio in real time. The article includes mathematical apparatus and technical implementation.

When exploring various model architecture designs, we often devote insufficient attention to the process of model training. In this article, I aim to address this gap.