Articles on machine learning in trading

​This article integrates the Optuna hyperparameter optimization (HPO) backend into a unified ModelDevelopmentPipeline. It adds joint tuning of model hyperparameters and sample-weight schemes, early pruning with Hyperband, and crash-resistant SQLite study storage. The pipeline auto-detects primary vs. secondary models, prepends a fitted column-dropping preprocessor for safe inference, supports sequential bootstrapping, generates an Optuna report, and includes bid/ask and LearnedStrategy links. Readers get faster, resumable runs and deployable, self-contained models.

We continue our last article on DDPG with MA and stochastic indicators by examining other key Reinforcement Learning classes crucial for implementing DDPG. Though we are mostly coding in python, the final product, of a trained network will be exported to as an ONNX to MQL5 where we integrate it as a resource in a wizard assembled Expert Advisor.

In this article, we will look at the Comet Tail Optimization Algorithm (CTA), which draws inspiration from unique space objects - comets and their impressive tails that form when approaching the Sun. The algorithm is based on the concept of the motion of comets and their tails, and is designed to find optimal solutions in optimization problems.

We continue to explore the analysis and forecasting of time series in the frequency domain. In this article, we will get acquainted with a new method to forecast data in the frequency domain, which can be added to many of the algorithms we have studied previously.

There are many posts in the MQL5 Forum asking for help calculating the slope of price changes. This article will demonstrate one possible way of calculating the angle formed by the changes in price in any market you wish to trade. Additionally, we will answer if engineering this new feature is worth the extra effort and time invested. We will explore if the slope of the price can improve any of our AI model's accuracy when forecasting the USDZAR pair on the M1.

The article presents a comprehensive analysis of the Coral Reef Optimization (CRO) algorithm, a metaheuristic method inspired by the biological processes of coral reef formation and development. The algorithm models key aspects of coral evolution: broadcast spawning, brooding, larval settlement, asexual reproduction, and competition for limited reef space. Particular attention is paid to the improved version of the algorithm.

In this article, we will continue exploring the Artificial Bee Hive Algorithm (ABHA) by diving into the code and considering the remaining methods. As you might remember, each bee in the model is represented as an individual agent whose behavior depends on internal and external information, as well as motivational state. We will test the algorithm on various functions and summarize the results by presenting them in the rating table.

In this article, we will get acquainted with an interesting algorithm that is built at the intersection of supervised and reinforcement learning methods.

Exploring advanced techniques to integrate MQL5 with powerful data processing tools, this part focuses on efficient handling of big data to enhance trading analysis and decision-making.

In this article, we will take a quick look at some methods to get a function that can represent our data in the database. I will not go into detail about how to use statistics and probability studies to interpret the results. Let's leave that for those who really want to delve into the mathematical side of the matter. Exploring these questions will be critical to understanding what is involved in studying neural networks. Here we will consider this issue quite calmly.

In the previous last article within this series, we looked at the Atom-Motif Contrastive Transformer (AMCT) framework, which uses contrastive learning to discover key patterns at all levels, from basic elements to complex structures. In this article, we continue implementing AMCT approaches using MQL5.

The article considers an interesting algorithm derived from inanimate nature - intelligent water drops (IWD) simulating the process of river bed formation. The ideas of this algorithm made it possible to significantly improve the previous leader of the rating - SDS. As usual, the new leader (modified SDSm) can be found in the attachment.

In a world overflowing with noisy and unpredictable data, identifying meaningful patterns can be challenging. In this article, we'll explore seasonal decomposition, a powerful analytical technique that helps separate data into its key components: trend, seasonal patterns, and noise. By breaking data down this way, we can uncover hidden insights and work with cleaner, more interpretable information.

Generative Adversarial Networks are a pairing of Neural Networks that train off of each other for more accurate results. We adopt the conditional type of these networks as we look to possible application in forecasting Financial time series within an Expert Signal Class.

We continue to study algorithms for extracting features from a point cloud. In this article, we will get acquainted with the mechanisms for increasing the efficiency of the PointNet method.

As the model is trained based on the experience reproduction buffer, the current Actor policy moves further and further away from the stored examples, which reduces the efficiency of training the model as a whole. In this article, we will look at the algorithm of improving the efficiency of using samples in reinforcement learning algorithms.

Dynamic mode decomposition (DMD) is a technique usually applied to high-dimensional datasets. In this article, we demonstrate the application of DMD on univariate time series, showing its ability to characterize a series as well as make forecasts. In doing so, we will investigate MQL5's built-in implementation of dynamic mode decomposition, paying particular attention to the new matrix method, DynamicModeDecomposition().

This series of articles introduces several time series labeling methods, which can create data that meets most artificial intelligence models, and targeted data labeling according to needs can make the trained artificial intelligence model more in line with the expected design, improve the accuracy of our model, and even help the model make a qualitative leap!

The article presents a complete Python–MQL5 integration for multi‑agent trading: MT5 data ingestion, indicator computation, per‑agent decisions, and a weighted consensus that outputs a single action. Signals are stored to JSON, served by Flask, and consumed by an MQL5 Expert Advisor for execution with position sizing and ATR‑derived SL/TP. Flask routes provide safe lifecycle control and status monitoring.

Number Walls are a variant of Linear Shift Back Registers that prescreen sequences for predictability by checking for convergence. We look at how these ideas could be of use in MQL5.

The article examines the impact of changing the shape of probability distributions on the performance of optimization algorithms. We will conduct experiments using the Smart Cephalopod (SC) test algorithm to evaluate the efficiency of various probability distributions in the context of optimization problems.

The article introduces the dialectical algorithm (DA), a new global optimization method inspired by the philosophical concept of dialectics. The algorithm exploits a unique division of the population into speculative and practical thinkers. Testing shows impressive performance of up to 98% on low-dimensional problems and overall efficiency of 57.95%. The article explains these metrics and presents a detailed description of the algorithm and the results of experiments on different types of functions.

Loss Function is the key metric of machine learning algorithms that provides feedback to the training process by quantifying how well a given set of parameters are performing when compared to their intended target. We explore the various formats of this function in an MQL5 custom wizard class.

We follow up on our last article, where we introduced the indicator pair of the Ichimoku and the ADX, by looking at how this duo could be improved with Supervised Learning. Ichimoku and ADX are a support/resistance plus trend complimentary pairing. Our supervised learning approach uses a neural network that engages the Deep Spectral Mixture Kernel to fine tune the forecasts of this indicator pairing. As per usual, this is done in a custom signal class file that works with the MQL5 wizard to assemble an Expert Advisor.

This is an improved chaotic optimization algorithm (COA) that combines the effects of chaos with adaptive search mechanisms. The algorithm uses a set of chaotic maps and inertial components to explore the search space. The article reveals the theoretical foundations of chaotic methods of financial optimization.

Moving Average and Stochastic Oscillator are very common indicators whose collective patterns we explored in the prior article, via a supervised learning network, to see which “patterns-would-stick”. We take our analyses from that article, a step further by considering the effects' reinforcement learning, when used with this trained network, would have on performance. Readers should note our testing is over a very limited time window. Nonetheless, we continue to harness the minimal coding requirements afforded by the MQL5 wizard in showcasing this.

We invite you to explore the ACEFormer architecture — a modern solution that combines the effectiveness of probabilistic attention with adaptive time series decomposition. This article will be useful for those seeking a balance between computational performance and forecast accuracy in financial markets.

We introduce CPyramidBridge, a thin MQL5 layer that maps bet-sizing results to CPyramidEngine. The bridge applies probability to initial lot sizing, enforces a capacity-aware entry gate, promotes add-ons from dynamic divergence, adapts the trailing stop to reserve estimates, and syncs signals on close, allowing an Expert Advisor to convert model confidence and concurrency into a structured, decreasing-risk pyramid.

The article considers a metaheuristic Artificial Ecosystem-based Optimization (AEO) algorithm, which simulates interactions between ecosystem components by creating an initial population of solutions and applying adaptive update strategies, and describes in detail the stages of AEO operation, including the consumption and decomposition phases, as well as different agent behavior strategies. The article introduces the features and advantages of this algorithm.

In this article, we revisit a classic crude oil trading strategy with the aim of enhancing it by leveraging supervised machine learning algorithms. We will construct a least-squares model to predict future Brent crude oil prices based on the spread between Brent and WTI crude oil prices. Our goal is to identify a leading indicator of future changes in Brent prices.

We continue studying the Hierarchical Vector Transformer method. In this article, we will complete the construction of the model. We will also train and test it on real historical data.

Reinforcement Learning is one of the three main tenets in machine learning, alongside supervised learning and unsupervised learning. It is therefore concerned with optimal control, or learning the best long-term policy that will best suit the objective function. It is with this back-drop, that we explore its possible role in informing the learning-process to an MLP of a wizard assembled Expert Advisor.

In the last article, we examined the pairing of Ichimoku and the ADX under an Inference Learning framework. For this piece we revisit, Reinforcement Learning when used with an indicator pairing we considered last in ‘Part 68’. The TRIX and Williams Percent Range. Our algorithm for this review will be the Quantile Regression DQN. As usual, we present this as a custom signal class designed for implementation with the MQL5 Wizard.

The article examines the Hamiltonian Monte Carlo (HMC) algorithm — the gold standard for sampling from complex multivariate distributions. A full-featured implementation of HMC in MQL5 is presented, including adaptive mass matrix tuning, MAP estimation using the L-BFGS optimization method, and comprehensive diagnostics.