Articles on machine learning in trading

In this article, we will get acquainted with an algorithm that uses closed-form policy improvement operators to optimize Agent actions in offline mode.

Biogeography-Based Optimization (BBO) is an elegant global optimization method inspired by natural processes of species migration between islands within archipelagos. The algorithm is based on a simple yet powerful idea: high-quality solutions actively share their characteristics, while low-quality ones actively adopt new features, creating a natural flow of information from the best solutions to the worst. A unique adaptive mutation operator provides an excellent balance between exploration and exploitation. BBO demonstrates high efficiency on a variety of tasks.

Kohonen's self-organizing maps transform the chaos of market data into an ordered two-dimensional map, where similar patterns are grouped together. The article demonstrates a complete implementation of a SOM in an MQL5 Expert Advisor with 400 neurons and continuous learning. We break down the Best Matching Unit search algorithm, weight updates using a Gaussian neighborhood function, integration with quantum effects, and the generation of trading signals. The code is open-source, the math is clear, and the results are verifiable.

The article considers the classification model of Gaussian processes. We will start by studying its theoretical principles moving on to the practical development of the GP library in MQL5.

We invite you to explore the original implementation of the K²VAE framework — a flexible model capable of linearly approximating complex dynamics in latent space. This article demonstrates how to implement key components in MQL5, including parameterized matrices and how to manage them outside standard neural network layers. This material will be useful for anyone looking for a practical approach to building interpretable time-series models.

The article discusses the projection of EURUSD daily candlesticks onto the secp256k1 elliptic curve and the extraction of 96 features (EC+TA) to predict the direction of the next candlestick in CatBoost. The price mapping to the curve and the training pipeline on 2,000 D1 bars are shown; the full model achieves a test AUC of 0.6508, with EC features contributing 60.6%. The materials are suitable for reproduction in Python/MetaTrader 5.

We invite you to explore the modern Time-MoE framework, which has been adapted for time series forecasting tasks. In this article, we will implement the key components of the architecture step by step, providing explanations and practical examples along the way. This approach will allow you not only to understand how the model works, but also to apply those principles to real-world trading scenarios.

The article presents a new approach to solving optimization problems by combining ideas from bacterial foraging optimization (BFO) algorithms and techniques used in the genetic algorithm (GA) into a hybrid BFO-GA algorithm. It uses bacterial swarming to globally search for an optimal solution and genetic operators to refine local optima. Unlike the original BFO, bacteria can now mutate and inherit genes.

We resume the topic of optimization by the scientific community. CoSO should not be viewed as a ready-made solution, but as a promising research platform. With proper development, CoSO can find its niche in tasks where adaptability and resilience to change are important, and computation time is not critical.

Building on the partition function analysis from Part 1, this article deepens the theoretical foundation before completing the analytical pipeline. We first give a full treatment of the Hurst exponent: what it measures, what it implies about market memory, and why it matters for the MMAR. This is followed by an intuitive exploration of multifractal spectra and what f(α) reveals about volatility heterogeneity. We then move to implementation: extracting the scaling function τ(q), estimating H via R/S analysis, and fitting the multifractal spectrum across four candidate distributions. By the end, we have the complete parameter set needed to construct the MMAR process in Part 3. Part 2 of an eight-part series.

In offline learning, we use a fixed dataset, which limits the coverage of environmental diversity. During the learning process, our Agent can generate actions beyond this dataset. If there is no feedback from the environment, how can we be sure that the assessments of such actions are correct? Maintaining the Agent's policy within the training dataset becomes an important aspect to ensure the reliability of training. This is what we will talk about in this article.

We are pleased to present the final part of our series on GinAR — a neural network framework for time series forecasting. In this article, we analyze the results of testing the model on new data and assess its robustness under real-market conditions.

This article starts the MMAR pipeline on EURUSD M5 data. We load market data via the MetaTrader5 Python API and run partition-function analysis with non-overlapping intervals to test for multifractal scaling. The result is an evidence-based decision on fractality, a prerequisite for building MMAR and for choosing whether to proceed beyond GARCH.

The article discusses the Cricket Algorithm, a metaheuristic optimization method that combines elements of the Bat Algorithm and the Firefly Algorithm with the physical laws governing the propagation of sound in the atmosphere. The algorithm simulates the behavior of crickets that navigate by the chirping of their conspecifics, using Dolbear's law and acoustic formulas to guide the search for best solutions.

This article builds a white-box classifier in MQL5 using the Tsetlin Machine. It learns human-readable AND-rules instead of weights, trains with integer state updates, and requires no external dependencies. You will assemble the automaton, clause, and multi-class voter, verify on XOR and other boolean tasks, booleanize indicators, label by forward ATR-scaled return, save the model to CSV, and view active rules on a live chart.

The article discusses the adaptation and practical implementation of the ACEFormer framework using MQL5 in the context of algorithmic trading. It presents key architectural decisions, training features, and model testing results on real data.

In this article we present the implementation of a feature selection algorithm described in an academic paper titled,"FREL: A stable feature selection algorithm", called Feature weighting as regularized energy based learning.

The article delves into the implementation of a modified Forward Selection Component Analysis algorithm, drawing inspiration from the research presented in “Forward Selection Component Analysis: Algorithms and Applications” by Luca Puggini and Sean McLoone.

The article presents a new metaheuristic method based on the hunting strategies of Australian dingoes: group attack, chase, and scavenging. Let's see how the Dingo Optimization Algorithm (DOA) performs algorithmically.

A revolutionary approach to machine learning in trading through quantum computing. The article demonstrates a practical implementation of an adaptive QRC system with continuous retraining for predicting market movements in real time.

Neural networks are already changing the way we analyze markets, and new architectures are opening up even more possibilities. In this article, we wrap up our work with the SpikingBrain framework, which opens up new possibilities for us.

In this article, we introduce a modified version of stepwise feature selection, implemented in MQL5. This approach is based on the techniques outlined in Modern Data Mining Algorithms in C++ and CUDA C by Timothy Masters.

We invite you to learn about the innovative technique of adaptive patching — a method for flexibly segmenting time series while taking their internal periodicity into account. We will also look at an efficient encoding technique that preserves important semantic characteristics when working with data at different scales. These methods open up new possibilities for the accurate processing of complex, multiscale data characteristic of financial markets and significantly improve the stability and reliability of forecasts.

XLV is SPDR healthcare ETF and in an age where it is common to be bombarded by a wide array of traditional news items plus social media feeds, it can be pressing to select a data set for use with a model. We try to tackle this problem for this ETF by sizing up some of its critical data sets in MQL5.

This article implements a Fenwick Tree (Binary Indexed Tree) for volume-aware money management inside an MQL5 Wizard Expert Advisor. We structure cumulative volume in O(log n) and apply four scaling modes—linear, conservative, aggressive, and mean-reversion—optionally gated by a lightweight 1D CNN. Practical tests compare the algorithm alone versus the CNN‑filtered approach to illustrate adaptive lot sizing and risk control under varying volume topologies.

This work presents an end-to-end pipeline: collect MetaTrader 5 data, engineer entropy/volatility/trend features, train a PyTorch classifier, and expose predictions through a Flask API. An MQL5 EA posts rolling prices each tick, receives probability and regime, and applies adaptive position sizing and stop distances. The result is a clear recipe for integrating ML inference with MetaTrader 5.

In this article we make the case for merging Divergence Mapping with a Temporal Fusion Proxy in a Trading Robot. Rather than depending on lagging price confirmations, the Divergence Mapping's thesis is that acting like a structural sensor can help identify hidden momentum shifts from price action and indicator anomalies. To establish how these anomalies are interpreted over time we use a Temporal Fusion Transformer proxy. This network incorporates historical context to weigh developing trends such that merging it with Divergence Mapping should set us up to spot shifts in accumulation and distribution before price breakouts.

A new optimization method, the Bison Algorithm (BIA), uses two strategies, inspired by the behavior of bison, for solving continuous problems with a single objective function. The key features of BIA are two fundamental principles borrowed from the behavior of bison: the ability to move dynamically and a defensive strategy.