Articles on machine learning in trading

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.

We invite you to explore the innovative SCNN framework, which takes time series analysis to a new level by clearly separating data into long-term, seasonal, short-term, and residual components. This approach significantly improves forecasting accuracy by allowing the model to adapt to complex and changing market dynamics.

This article presents a complete RL trading pipeline for XAUUSD: a supervised signal baseline with triple-barrier labels, PPO training, purged walk-forward validation with embargo, multi-seed checks, and contract-guarded deployment with normalization. It includes runnable code for data validation, features, environment, training, and broker‑based reconciliation. The live demo over 763 closed trades showed no statistically significant edge, and the methods highlight where information and costs, not architecture, set performance limits.

The Dandelion Optimizer (DO) turns the simple flight of a seed carried by the wind into a mathematical search strategy. The three phases — vortex rising, drift toward the center of the population, and landing along a Lévy-flight trajectory — form an elegant metaphor that yields interesting results in practice.

Raw feature correlations contain estimation noise and a shared market-mode component that distort clustering. We fit the Marcenko–Pastur noise ceiling (with an effective sample size correction), apply constant-residual denoising and market detonation, and run the Optimal Number of Clusters routine. The result is a cleaned correlation matrix and stable cluster labels that avoid substitution effects and feed clustered MDI/MDA in the next 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.

The Dendritic Cell Algorithm (DCA) is a metaheuristic inspired by the mechanisms of the innate immune system. Dendritic cells patrol the search space, accumulate signals about the quality of positions, and reach a collective decision: whether to exploit what they have found or to continue exploration. Let's take a look at how a biological model for detecting pathogens is transformed into an optimization algorithm.

This article implements a self-contained Isolation Forest library for MetaTrader 5 with no labels, no distribution assumptions and no external dependencies. It details a reproducible 64‑bit generator, tree/forest construction, scoring and feature design, then verifies results against Python and market data with two null models. The package includes an indicator that plots the decision variable and a gate example. Readers get a validated library, clear limits of applicability and a practical way to calibrate thresholds.

We invite you to explore a new implementation of the key components of the GinAR framework — an adaptive algorithm for working with graph-structured time series. This article provides a step-by-step breakdown of the architecture and the algorithms for the forward pass and error backpropagation.

The article presents an adaptation of the Deterministic Dendritic Cell Algorithm (dDCA) for continuous optimization problems. The algorithm, inspired by the immune system's Danger Theory, uses a signal accumulation mechanism to automatically balance exploration and exploitation within the search space.

The article describes the practical implementation of a hybrid algorithmic trading system that combines quantum computing (IBM Qiskit) and gradient boosting (CatBoost) to predict movements in the EURUSD pair on the hourly time frame. The system extracts four unique quantum features from a probability distribution across 256 states using eight qubits and, in combination with classical indicators and delta encoding of time categories, achieves 62% accuracy on 15,000 candlesticks.

We build the Matrix Profile for MQL5 from the ground up and keep it numerically stable on real prices. The library includes rolling statistics, a radix-2 FFT powering MASS, and a STOMP self-join, with results matched to stumpy. A compact facade, an indicator that draws the profile and flags discords, and a demonstration Expert Advisor show how to read and use the signal in practice.

We invite you to learn about the K²VAE framework and how the proposed approaches can be integrated into a trading system. You will learn how the hybrid Koopman–Kalman–VAE approach helps build adaptive and interpretable models. The article concludes with practical results from using the implemented solutions.

This article presents a compact MQL5 library for market-complexity analysis: LZ76 complexity and Normalized Compression Distance built on a SAX symbolizer, exposed through a simple facade and an efficiency indicator. It explains the discretization choices, normalization, and distance formulation, and validates the code with unit checks and an independent cross-check. You get a ready-to-use library and indicator, plus a disciplined way to interpret readings with a shuffle null and a direction check.

This article implements an online logistic‑regression trade filter in native MQL5 and integrates it into an EMA‑crossover EA with a closed‑trade feedback loop. It details the shared class, features, SGD update, persistence, and a read‑only probability view. Synthetic experiments cover multi‑seed separation, calibration, feature ablation, regime‑shift baselines, and hyperparameter sweeps. You get reproducible scripts and a walk‑forward protocol to validate the filter on your own instrument.

Price outliers distort indicators based on the mean and standard deviation. This article delivers a robust MQL5 library (RobustStats.mqh) implementing the median, 1.4826-scaled MAD, and Theil–Sen slope, plus three drop‑in indicators that replace Bollinger Bands, the linear regression channel, and the z‑score oscillator. A comparison overlay and a breakdown‑point measurement on EURUSD show how the robust instruments hold their shape when a single spike moves the classical ones.

We invite you to explore a new approach that combines classical methods and modern neural networks for time series analysis. The article provides a detailed explanation of the architecture and operating principles of the K²VAE model.

We evaluate blind source separation for market noise control using FastICA applied to SMA-filtered, time-lagged OHLC features. The study compares classical and surrogate targets, measures accuracy across lags, tunes KNN models, and inspects residual structure with clustering. Models are exported to ONNX and integrated into an MQL5 Expert Advisor for testing. The result is a reproducible pipeline from data extraction to deployment.

The article discusses the Differential Search Algorithm (DSA), which simulates the migration of a superorganism in search of optimal living conditions. The algorithm uses a Gamma distribution to generate a pseudo-stable random walk and offers four strategies for selecting the direction of movement, along with three coordinate mutation mechanisms. How will this method perform?

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.

We complete the native MQL5 port of Kronos: the decoder, the predictor's decode_s1 and decode_s2 stages with their cross-attention traps, and the autoregressive loop that produces a multi-bar forecast. Then we profile and make it roughly 4.5x faster with an exact KV-cache and pre-transposed weights, verifying every stage against PyTorch.

Trend-scanning supports both forward and backward windows, and the labeling default is unsafe for features: it looks ahead and boosts next-bar agreement well above chance on random walks. We provide a dedicated wrapper, get trend scanning features, that forces computational causal and returns only window, slope, t value, and rsquared. A second analysis quantifies errors introduced by the default log transform on signed series.

We invite you to dive into the exciting world of LightGTS — a lightweight yet powerful framework for time-series forecasting, where adaptive convolution and RoPE encoding are combined with innovative attention mechanisms. In our article, you will find a detailed description of all components — from creating patches to the complex mixture of experts in the decoder — ready for integration into MQL5 projects. Discover how LightGTS takes automated trading to a whole new level!

The article shows how to evaluate machine-learning alphas before a full backtest by expressing them as formulaic alphas. We compute Information Coefficient (IC), Rank IC, Information Ratio (ICIR), and t-statistics to quantify forecasting strength and stability. A MetaTrader 5 backtest illustrates differences versus execution-dependent tests, and a Python parser facilitates reproducible calculations and bulk screening.

This article builds a Kolmogorov–Arnold Network (KAN) in MQL5, where every edge carries a learnable B‑spline curve rather than a scalar weight. We construct the spline basis, assemble edges and a layer, and fit all coefficients by ridge‑regularized least‑squares in a single solve. The model is delivered as an indicator that visualizes the learned curves and an Expert Advisor that acts on the prediction, providing an interpretable, reusable codebase.

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.

Experimental evaluation on standard benchmark functions reveals the advantages and limitations of directly adapting combinatorial algorithms. The article provides a detailed description of the ECEA algorithm's mechanisms and test results.

This article performs a numerical verification of MQL5 eigendecomposition for a covariance matrix using the spectral theorem A = V Λ Vᵀ. It reconstructs the matrix with Diag(), Transpose(), and MatMul(), computes the residual and its Frobenius norm, and shows that deviations remain at floating‑point precision, with results printed to the Experts journal.

The article provides a fascinating look at how SwiGLU embedding reveals hidden market patterns, and how a sparse Mixture of Experts within a Decoder-Only Transformer makes forecasts more accurate at reasonable computational cost. We take an in-depth look at the integration of Time‑MoE into MQL5 and OpenCL, and provide a step-by-step guide to configuring and training the model.

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.

We invite you to explore the practical implementation of a sparse mixture of experts block for time series in the OpenCL computing environment. This article provides a step-by-step explanation of how masked multi-window convolution works, as well as how gradient-based training is organized in the presence of multiple information streams.